src/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_template_avx_128_fma_single.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_avx_128_fma_single.h"
  17 #include "kernelutil_x86_avx_128_fma_single.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,C,D refer to j loop unrolling done with AVX_128, e.g. for the four 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,jnrC,jnrD;
  80     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  81     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  86     real             scratch[4*DIM];
  87     __m128           fscal,rcutoff,rcutoff2,jidxall;
  88     /* #for I in PARTICLES_I */
  89     int              vdwioffset{I};
  90     __m128           ix{I},iy{I},iz{I},fix{I},fiy{I},fiz{I},iq{I},isai{I};
  91     /* #endfor */
  92     /* #for J in PARTICLES_J */
  93     int              vdwjidx{J}A,vdwjidx{J}B,vdwjidx{J}C,vdwjidx{J}D;
  94     __m128           jx{J},jy{J},jz{J},fjx{J},fjy{J},fjz{J},jq{J},isaj{J};
  95     /* #endfor */
  96     /* #for I,J in PAIRS_IJ */
  97     __m128           dx{I}{J},dy{I}{J},dz{I}{J},rsq{I}{J},rinv{I}{J},rinvsq{I}{J},r{I}{J},qq{I}{J},c6_{I}{J},c12_{I}{J};
  98     /* #endfor */
  99     /* #if KERNEL_ELEC != 'None' */
 100     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 101     real             *charge;
 102     /* #endif */
 103     /* #if 'GeneralizedBorn' in KERNEL_ELEC */
 104     __m128i          gbitab;
 105     __m128           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,twogbeps,dvdatmp;
 106     __m128           minushalf = _mm_set1_ps(-0.5);
 107     real             *invsqrta,*dvda,*gbtab;
 108     /* #endif */
 109     /* #if KERNEL_VDW != 'None' */
 110     int              nvdwtype;
 111     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 112     int              *vdwtype;
 113     real             *vdwparam;
 114     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 115     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 116     /* #endif */
 117     /* #if 'Table' in KERNEL_ELEC or 'GeneralizedBorn' in KERNEL_ELEC or 'Table' in KERNEL_VDW */
 118     __m128i          vfitab;
 119     __m128i          ifour       = _mm_set1_epi32(4);
 120     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 121     real             *vftab;
 122     /* #endif */
 123     /* #if 'Ewald' in KERNEL_ELEC */
 124     __m128i          ewitab;
 125     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 126     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 127     real             *ewtab;
 128     /* #endif */
 129     /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
 130     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 131     real             rswitch_scalar,d_scalar;
 132     /* #endif */
 133     __m128           dummy_mask,cutoff_mask;
 134     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 135     __m128           one     = _mm_set1_ps(1.0);
 136     __m128           two     = _mm_set1_ps(2.0);
 137     x                = xx[0];
 138     f                = ff[0];
 139
 140     nri              = nlist->nri;
 141     iinr             = nlist->iinr;
 142     jindex           = nlist->jindex;
 143     jjnr             = nlist->jjnr;
 144     shiftidx         = nlist->shift;
 145     gid              = nlist->gid;
 146     shiftvec         = fr->shift_vec[0];
 147     fshift           = fr->fshift[0];
 148     /* #if KERNEL_ELEC != 'None' */
 149     facel            = _mm_set1_ps(fr->epsfac);
 150     charge           = mdatoms->chargeA;
 151     /*     #if 'ReactionField' in KERNEL_ELEC */
 152     krf              = _mm_set1_ps(fr->ic->k_rf);
 153     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 154     crf              = _mm_set1_ps(fr->ic->c_rf);
 155     /*     #endif */
 156     /* #endif */
 157     /* #if KERNEL_VDW != 'None' */
 158     nvdwtype         = fr->ntype;
 159     vdwparam         = fr->nbfp;
 160     vdwtype          = mdatoms->typeA;
 161     /* #endif */
 162
 163     /* #if 'Table' in KERNEL_ELEC and 'Table' in KERNEL_VDW */
 164     vftab            = kernel_data->table_elec_vdw->data;
 165     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 166     /* #elif 'Table' in KERNEL_ELEC */
 167     vftab            = kernel_data->table_elec->data;
 168     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 169     /* #elif 'Table' in KERNEL_VDW */
 170     vftab            = kernel_data->table_vdw->data;
 171     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 172     /* #endif */
 173
 174     /* #if 'Ewald' in KERNEL_ELEC */
 175     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 176     beta             = _mm_set1_ps(fr->ic->ewaldcoeff);
 177     beta2            = _mm_mul_ps(beta,beta);
 178     beta3            = _mm_mul_ps(beta,beta2);
 179     /*     #if KERNEL_VF=='Force' and KERNEL_MOD_ELEC!='PotentialSwitch' */
 180     ewtab            = fr->ic->tabq_coul_F;
 181     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 182     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 183     /*     #else */
 184     ewtab            = fr->ic->tabq_coul_FDV0;
 185     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 186     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 187      /*     #endif */
 188     /* #endif */
 189
 190     /* #if KERNEL_ELEC=='GeneralizedBorn' */
 191     invsqrta         = fr->invsqrta;
 192     dvda             = fr->dvda;
 193     gbtabscale       = _mm_set1_ps(fr->gbtab.scale);
 194     gbtab            = fr->gbtab.data;
 195     gbinvepsdiff     = _mm_set1_ps((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 196     /* #endif */
 197
 198     /* #if 'Water' in GEOMETRY_I */
 199     /* Setup water-specific parameters */
 200     inr              = nlist->iinr[0];
 201     /*     #for I in PARTICLES_ELEC_I */
 202     iq{I}              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+{I}]));
 203     /*     #endfor */
 204     /*     #for I in PARTICLES_VDW_I */
 205     vdwioffset{I}      = 2*nvdwtype*vdwtype[inr+{I}];
 206     /*     #endfor */
 207     /* #endif */
 208
 209     /* #if 'Water' in GEOMETRY_J */
 210     /*     #for J in PARTICLES_ELEC_J */
 211     jq{J}              = _mm_set1_ps(charge[inr+{J}]);
 212     /*     #endfor */
 213     /*     #for J in PARTICLES_VDW_J */
 214     vdwjidx{J}A        = 2*vdwtype[inr+{J}];
 215     /*     #endfor */
 216     /*     #for I,J in PAIRS_IJ */
 217     /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 218     qq{I}{J}             = _mm_mul_ps(iq{I},jq{J});
 219     /*         #endif */
 220     /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 221     c6_{I}{J}            = _mm_set1_ps(vdwparam[vdwioffset{I}+vdwjidx{J}A]);
 222     c12_{I}{J}           = _mm_set1_ps(vdwparam[vdwioffset{I}+vdwjidx{J}A+1]);
 223     /*         #endif */
 224     /*     #endfor */
 225     /* #endif */
 226
 227     /* #if KERNEL_MOD_ELEC!='None' or KERNEL_MOD_VDW!='None' */
 228     /*     #if KERNEL_ELEC!='None' */
 229     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 230     rcutoff_scalar   = fr->rcoulomb;
 231     /*     #else */
 232     rcutoff_scalar   = fr->rvdw;
 233     /*     #endif */
 234     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 235     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 236     /* #endif */
 237
 238     /* #if KERNEL_MOD_VDW=='PotentialShift' */
 239     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 240     rvdw             = _mm_set1_ps(fr->rvdw);
 241     /* #endif */
 242
 243     /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
 244     /*     #if KERNEL_MOD_ELEC=='PotentialSwitch'  */
 245     rswitch_scalar   = fr->rcoulomb_switch;
 246     rswitch          = _mm_set1_ps(rswitch_scalar);
 247     /*     #else */
 248     rswitch_scalar   = fr->rvdw_switch;
 249     rswitch          = _mm_set1_ps(rswitch_scalar);
 250     /*     #endif */
 251     /* Setup switch parameters */
 252     d_scalar         = rcutoff_scalar-rswitch_scalar;
 253     d                = _mm_set1_ps(d_scalar);
 254     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 255     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 256     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 257     /*     #if 'Force' in KERNEL_VF */
 258     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 259     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 260     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 261     /*     #endif */
 262     /* #endif */
 263
 264     /* Avoid stupid compiler warnings */
 265     jnrA = jnrB = jnrC = jnrD = 0;
 266     j_coord_offsetA = 0;
 267     j_coord_offsetB = 0;
 268     j_coord_offsetC = 0;
 269     j_coord_offsetD = 0;
 270
 271     /* ## Keep track of the floating point operations we issue for reporting! */
 272     /* #define OUTERFLOPS 0 */
 273     outeriter        = 0;
 274     inneriter        = 0;
 275
 276     for(iidx=0;iidx<4*DIM;iidx++)
 277     {
 278         scratch[iidx] = 0.0;
 279     }
 280
 281     /* Start outer loop over neighborlists */
 282     for(iidx=0; iidx<nri; iidx++)
 283     {
 284         /* Load shift vector for this list */
 285         i_shift_offset   = DIM*shiftidx[iidx];
 286
 287         /* Load limits for loop over neighbors */
 288         j_index_start    = jindex[iidx];
 289         j_index_end      = jindex[iidx+1];
 290
 291         /* Get outer coordinate index */
 292         inr              = iinr[iidx];
 293         i_coord_offset   = DIM*inr;
 294
 295         /* Load i particle coords and add shift vector */
 296         /* #if GEOMETRY_I == 'Particle' */
 297         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 298         /* #elif GEOMETRY_I == 'Water3' */
 299         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 300                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 301         /* #elif GEOMETRY_I == 'Water4' */
 302         /*     #if 0 in PARTICLES_I                 */
 303         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 304                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 305         /*     #else                                */
 306         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 307                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 308         /*     #endif                               */
 309         /* #endif                                   */
 310
 311         /* #if 'Force' in KERNEL_VF */
 312         /*     #for I in PARTICLES_I */
 313         fix{I}             = _mm_setzero_ps();
 314         fiy{I}             = _mm_setzero_ps();
 315         fiz{I}             = _mm_setzero_ps();
 316         /*     #endfor */
 317         /* #endif */
 318
 319         /* ## For water we already preloaded parameters at the start of the kernel */
 320         /* #if not 'Water' in GEOMETRY_I */
 321         /* Load parameters for i particles */
 322         /*     #for I in PARTICLES_ELEC_I */
 323         iq{I}              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+{I}));
 324         /*         #define OUTERFLOPS OUTERFLOPS+1 */
 325         /*         #if KERNEL_ELEC=='GeneralizedBorn' */
 326         isai{I}            = _mm_load1_ps(invsqrta+inr+{I});
 327         /*         #endif */
 328         /*     #endfor */
 329         /*     #for I in PARTICLES_VDW_I */
 330         vdwioffset{I}      = 2*nvdwtype*vdwtype[inr+{I}];
 331         /*     #endfor */
 332         /* #endif */
 333
 334         /* #if 'Potential' in KERNEL_VF */
 335         /* Reset potential sums */
 336         /*     #if KERNEL_ELEC != 'None' */
 337         velecsum         = _mm_setzero_ps();
 338         /*     #endif */
 339         /*     #if 'GeneralizedBorn' in KERNEL_ELEC */
 340         vgbsum           = _mm_setzero_ps();
 341         /*     #endif */
 342         /*     #if KERNEL_VDW != 'None' */
 343         vvdwsum          = _mm_setzero_ps();
 344         /*     #endif */
 345         /* #endif */
 346         /*     #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
 347         dvdasum          = _mm_setzero_ps();
 348         /*     #endif */
 349
 350         /* #for ROUND in ['Loop','Epilogue'] */
 351
 352         /* #if ROUND =='Loop' */
 353         /* Start inner kernel loop */
 354         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 355         {
 356         /* ## First round is normal loop (next statement resets indentation) */
 357         /*     #if 0 */
 358         }
 359         /*     #endif */
 360         /* #else */
 361         if(jidx<j_index_end)
 362         {
 363         /* ## Second round is epilogue */
 364         /* #endif */
 365         /* #define INNERFLOPS 0 */
 366
 367             /* Get j neighbor index, and coordinate index */
 368             /* #if ROUND =='Loop' */
 369             jnrA             = jjnr[jidx];
 370             jnrB             = jjnr[jidx+1];
 371             jnrC             = jjnr[jidx+2];
 372             jnrD             = jjnr[jidx+3];
 373             /* #else */
 374             jnrlistA         = jjnr[jidx];
 375             jnrlistB         = jjnr[jidx+1];
 376             jnrlistC         = jjnr[jidx+2];
 377             jnrlistD         = jjnr[jidx+3];
 378             /* Sign of each element will be negative for non-real atoms.
 379              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 380              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 381              */
 382             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 383             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 384             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 385             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 386             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 387             /* #endif */
 388             j_coord_offsetA  = DIM*jnrA;
 389             j_coord_offsetB  = DIM*jnrB;
 390             j_coord_offsetC  = DIM*jnrC;
 391             j_coord_offsetD  = DIM*jnrD;
 392
 393             /* load j atom coordinates */
 394             /* #if GEOMETRY_J == 'Particle'             */
 395             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 396                                               x+j_coord_offsetC,x+j_coord_offsetD,
 397                                               &jx0,&jy0,&jz0);
 398             /* #elif GEOMETRY_J == 'Water3'             */
 399             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 400                                               x+j_coord_offsetC,x+j_coord_offsetD,
 401                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 402             /* #elif GEOMETRY_J == 'Water4'             */
 403             /*     #if 0 in PARTICLES_J                 */
 404             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 405                                               x+j_coord_offsetC,x+j_coord_offsetD,
 406                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 407                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 408             /*     #else                                */
 409             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 410                                               x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 411                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 412             /*     #endif                               */
 413             /* #endif                                   */
 414
 415             /* Calculate displacement vector */
 416             /* #for I,J in PAIRS_IJ */
 417             dx{I}{J}             = _mm_sub_ps(ix{I},jx{J});
 418             dy{I}{J}             = _mm_sub_ps(iy{I},jy{J});
 419             dz{I}{J}             = _mm_sub_ps(iz{I},jz{J});
 420             /*     #define INNERFLOPS INNERFLOPS+3 */
 421             /* #endfor */
 422
 423             /* Calculate squared distance and things based on it */
 424             /* #for I,J in PAIRS_IJ */
 425             rsq{I}{J}            = gmx_mm_calc_rsq_ps(dx{I}{J},dy{I}{J},dz{I}{J});
 426             /*     #define INNERFLOPS INNERFLOPS+5 */
 427             /* #endfor */
 428
 429             /* #for I,J in PAIRS_IJ */
 430             /*     #if 'rinv' in INTERACTION_FLAGS[I][J] */
 431             rinv{I}{J}           = gmx_mm_invsqrt_ps(rsq{I}{J});
 432             /*         #define INNERFLOPS INNERFLOPS+5 */
 433             /*     #endif */
 434             /* #endfor */
 435
 436             /* #for I,J in PAIRS_IJ */
 437             /*     #if 'rinvsq' in INTERACTION_FLAGS[I][J] */
 438             /*         # if 'rinv' not in INTERACTION_FLAGS[I][J] */
 439             rinvsq{I}{J}         = gmx_mm_inv_ps(rsq{I}{J});
 440             /*             #define INNERFLOPS INNERFLOPS+4 */
 441             /*         #else */
 442             rinvsq{I}{J}         = _mm_mul_ps(rinv{I}{J},rinv{I}{J});
 443             /*             #define INNERFLOPS INNERFLOPS+1 */
 444             /*         #endif */
 445             /*     #endif */
 446             /* #endfor */
 447
 448             /* #if not 'Water' in GEOMETRY_J */
 449             /* Load parameters for j particles */
 450             /*     #for J in PARTICLES_ELEC_J */
 451             jq{J}              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+{J},charge+jnrB+{J},
 452                                                               charge+jnrC+{J},charge+jnrD+{J});
 453             /*         #if KERNEL_ELEC=='GeneralizedBorn' */
 454             isaj{J}            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+{J},invsqrta+jnrB+{J},
 455                                                               invsqrta+jnrC+{J},invsqrta+jnrD+{J});
 456             /*         #endif */
 457             /*     #endfor */
 458             /*     #for J in PARTICLES_VDW_J */
 459             vdwjidx{J}A        = 2*vdwtype[jnrA+{J}];
 460             vdwjidx{J}B        = 2*vdwtype[jnrB+{J}];
 461             vdwjidx{J}C        = 2*vdwtype[jnrC+{J}];
 462             vdwjidx{J}D        = 2*vdwtype[jnrD+{J}];
 463             /*     #endfor */
 464             /* #endif */
 465
 466             /* #if 'Force' in KERNEL_VF and not 'Particle' in GEOMETRY_I */
 467             /*     #for J in PARTICLES_J */
 468             fjx{J}             = _mm_setzero_ps();
 469             fjy{J}             = _mm_setzero_ps();
 470             fjz{J}             = _mm_setzero_ps();
 471             /*     #endfor */
 472             /* #endif */
 473
 474             /* #for I,J in PAIRS_IJ */
 475
 476             /**************************
 477              * CALCULATE INTERACTIONS *
 478              **************************/
 479
 480             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 481             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 482             /*         ## We always calculate rinv/rinvsq above to enable pipelineing in compilers (performance tested on x86) */
 483             if (gmx_mm_any_lt(rsq{I}{J},rcutoff2))
 484             {
 485                 /*     #if 0    ## this and the next two lines is a hack to maintain auto-indentation in template file */
 486             }
 487             /*         #endif */
 488             /*         #define INNERFLOPS INNERFLOPS+1 */
 489             /*     #endif */
 490
 491             /*     #if 'r' in INTERACTION_FLAGS[I][J] */
 492             r{I}{J}              = _mm_mul_ps(rsq{I}{J},rinv{I}{J});
 493             /*         #if ROUND == 'Epilogue' */
 494             r{I}{J}              = _mm_andnot_ps(dummy_mask,r{I}{J});
 495             /*             #define INNERFLOPS INNERFLOPS+1 */
 496             /*         #endif */
 497             /*         #define INNERFLOPS INNERFLOPS+1 */
 498             /*     #endif */
 499
 500             /*     ## For water geometries we already loaded parameters at the start of the kernel */
 501             /*     #if not 'Water' in GEOMETRY_J */
 502             /* Compute parameters for interactions between i and j atoms */
 503             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 504             qq{I}{J}             = _mm_mul_ps(iq{I},jq{J});
 505             /*             #define INNERFLOPS INNERFLOPS+1 */
 506             /*         #endif */
 507             /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 508             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset{I}+vdwjidx{J}A,
 509                                          vdwparam+vdwioffset{I}+vdwjidx{J}B,
 510                                          vdwparam+vdwioffset{I}+vdwjidx{J}C,
 511                                          vdwparam+vdwioffset{I}+vdwjidx{J}D,
 512                                          &c6_{I}{J},&c12_{I}{J});
 513             /*         #endif */
 514             /*     #endif */
 515
 516             /*     #if 'table' in INTERACTION_FLAGS[I][J] */
 517             /* Calculate table index by multiplying r with table scale and truncate to integer */
 518             rt               = _mm_mul_ps(r{I}{J},vftabscale);
 519             vfitab           = _mm_cvttps_epi32(rt);
 520 #ifdef __XOP__
 521             vfeps            = _mm_frcz_ps(rt);
 522 #else
 523             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 524 #endif
 525             twovfeps         = _mm_add_ps(vfeps,vfeps);
 526             /*         #define INNERFLOPS INNERFLOPS+4                          */
 527             /*         #if 'Table' in KERNEL_ELEC and 'Table' in KERNEL_VDW     */
 528             /*             ## 3 tables, 4 bytes per point: multiply index by 12 */
 529             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 530             /*         #elif 'Table' in KERNEL_ELEC                             */
 531             /*             ## 1 table, 4 bytes per point: multiply index by 4   */
 532             vfitab           = _mm_slli_epi32(vfitab,2);
 533             /*         #elif 'Table' in KERNEL_VDW                              */
 534             /*             ## 2 tables, 4 bytes per point: multiply index by 8  */
 535             vfitab           = _mm_slli_epi32(vfitab,3);
 536             /*         #endif                                                   */
 537             /*     #endif */
 538
 539             /*     ## ELECTROSTATIC INTERACTIONS */
 540             /*     #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 541
 542             /*         #if KERNEL_ELEC=='Coulomb' */
 543
 544             /* COULOMB ELECTROSTATICS */
 545             velec            = _mm_mul_ps(qq{I}{J},rinv{I}{J});
 546             /*             #define INNERFLOPS INNERFLOPS+1 */
 547             /*             #if 'Force' in KERNEL_VF */
 548             felec            = _mm_mul_ps(velec,rinvsq{I}{J});
 549             /*                 #define INNERFLOPS INNERFLOPS+2 */
 550             /*             #endif */
 551
 552             /*         #elif KERNEL_ELEC=='ReactionField' */
 553
 554             /* REACTION-FIELD ELECTROSTATICS */
 555             /*             #if 'Potential' in KERNEL_VF */
 556             velec            = _mm_mul_ps(qq{I}{J},_mm_sub_ps(_mm_macc_ps(krf,rsq{I}{J},rinv{I}{J}),crf));
 557             /*                 #define INNERFLOPS INNERFLOPS+4 */
 558             /*             #endif */
 559             /*             #if 'Force' in KERNEL_VF */
 560             felec            = _mm_mul_ps(qq{I}{J},_mm_msub_ps(rinv{I}{J},rinvsq{I}{J},krf2));
 561             /*                 #define INNERFLOPS INNERFLOPS+3 */
 562             /*             #endif */
 563
 564             /*         #elif KERNEL_ELEC=='GeneralizedBorn' */
 565
 566             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 567             isaprod          = _mm_mul_ps(isai{I},isaj{J});
 568             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq{I}{J},_mm_mul_ps(isaprod,gbinvepsdiff)));
 569             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 570             /*             #define INNERFLOPS INNERFLOPS+5 */
 571
 572             /* Calculate generalized born table index - this is a separate table from the normal one,
 573              * but we use the same procedure by multiplying r with scale and truncating to integer.
 574              */
 575             rt               = _mm_mul_ps(r{I}{J},gbscale);
 576             gbitab           = _mm_cvttps_epi32(rt);
 577 #ifdef __XOP__
 578             gbeps            = _mm_frcz_ps(rt);
 579 #else
 580             gbeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 581 #endif
 582             gbitab           = _mm_slli_epi32(gbitab,2);
 583
 584             Y                = _mm_load_ps( gbtab + _mm_extract_epi32(gbitab,0) );
 585             F                = _mm_load_ps( gbtab + _mm_extract_epi32(gbitab,1) );
 586             G                = _mm_load_ps( gbtab + _mm_extract_epi32(gbitab,2) );
 587             H                = _mm_load_ps( gbtab + _mm_extract_epi32(gbitab,3) );
 588             _MM_TRANSPOSE4_PS(Y,F,G,H);
 589             Fp               = _mm_macc_ps(gbeps,_mm_macc_ps(gbeps,H,G),F);
 590             VV               = _mm_macc_ps(gbeps,Fp,Y);
 591             vgb              = _mm_mul_ps(gbqqfactor,VV);
 592             /*             #define INNERFLOPS INNERFLOPS+10 */
 593
 594             /*             #if 'Force' in KERNEL_VF */
 595             twogbeps         = _mm_add_ps(gbeps,gbeps);
 596             FF               = _mm_macc_ps(_mm_macc_ps(twogbeps,H,G),gbeps,Fp);
 597             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 598             dvdatmp          = _mm_mul_ps(minushalf,_mm_macc_ps(fgb,r{I}{J},vgb));
 599             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 600             /*                 #if ROUND == 'Loop' */
 601             fjptrA           = dvda+jnrA;
 602             fjptrB           = dvda+jnrB;
 603             fjptrC           = dvda+jnrC;
 604             fjptrD           = dvda+jnrD;
 605             /*                 #else */
 606             /* The pointers to scratch make sure that this code with compilers that take gmx_restrict seriously (e.g. icc 13) really can't screw things up. */
 607             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 608             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 609             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 610             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 611             /*                 #endif */
 612             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj{J},isaj{J})));
 613             /*                 #define INNERFLOPS INNERFLOPS+13 */
 614             /*             #endif */
 615             velec            = _mm_mul_ps(qq{I}{J},rinv{I}{J});
 616             /*                 #define INNERFLOPS INNERFLOPS+1 */
 617             /*             #if 'Force' in KERNEL_VF */
 618             felec            = _mm_mul_ps(_mm_msub_ps(velec,rinv{I}{J},fgb),rinv{I}{J});
 619             /*                 #define INNERFLOPS INNERFLOPS+3 */
 620             /*             #endif */
 621
 622             /*         #elif KERNEL_ELEC=='Ewald' */
 623             /* EWALD ELECTROSTATICS */
 624
 625             /* Analytical PME correction */
 626             zeta2            = _mm_mul_ps(beta2,rsq{I}{J});
 627             /*             #if 'Force' in KERNEL_VF */
 628             rinv3            = _mm_mul_ps(rinvsq{I}{J},rinv{I}{J});
 629             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 630             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 631             felec            = _mm_mul_ps(qq{I}{J},felec);
 632             /*             #endif */
 633             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_ELEC=='PotentialSwitch' */
 634             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 635             /*                 #if KERNEL_MOD_ELEC=='PotentialShift' */
 636             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv{I}{J},sh_ewald));
 637             /*                 #else */
 638             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv{I}{J});
 639             /*                 #endif */
 640             velec            = _mm_mul_ps(qq{I}{J},velec);
 641             /*             #endif */
 642
 643             /*         #elif KERNEL_ELEC=='CubicSplineTable' */
 644
 645             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 646             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 647             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 648             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 649             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 650             _MM_TRANSPOSE4_PS(Y,F,G,H);
 651             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 652             /*             #define INNERFLOPS INNERFLOPS+4 */
 653             /*             #if 'Potential' in KERNEL_VF */
 654             VV               = _mm_macc_ps(vfeps,Fp,Y);
 655             velec            = _mm_mul_ps(qq{I}{J},VV);
 656             /*                 #define INNERFLOPS INNERFLOPS+3 */
 657             /*             #endif */
 658             /*             #if 'Force' in KERNEL_VF */
 659             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 660             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq{I}{J},FF),_mm_mul_ps(vftabscale,rinv{I}{J})));
 661             /*                 #define INNERFLOPS INNERFLOPS+7 */
 662             /*             #endif */
 663             /*         #endif */
 664             /*         ## End of check for electrostatics interaction forms */
 665             /*     #endif */
 666             /*     ## END OF ELECTROSTATIC INTERACTION CHECK FOR PAIR I-J */
 667
 668             /*     #if 'vdw' in INTERACTION_FLAGS[I][J] */
 669
 670             /*         #if KERNEL_VDW=='LennardJones' */
 671
 672             /* LENNARD-JONES DISPERSION/REPULSION */
 673
 674             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq{I}{J},rinvsq{I}{J}),rinvsq{I}{J});
 675             /*             #define INNERFLOPS INNERFLOPS+2 */
 676             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_VDW=='PotentialSwitch' */
 677             vvdw6            = _mm_mul_ps(c6_{I}{J},rinvsix);
 678             vvdw12           = _mm_mul_ps(c12_{I}{J},_mm_mul_ps(rinvsix,rinvsix));
 679             /*                 #define INNERFLOPS INNERFLOPS+3 */
 680             /*                 #if KERNEL_MOD_VDW=='PotentialShift' */
 681             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_{I}{J},_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 682                                           _mm_mul_ps( _mm_nmacc_ps(c6_{I}{J},sh_vdw_invrcut6,vvdw6),one_sixth));
 683             /*                     #define INNERFLOPS INNERFLOPS+8 */
 684             /*                 #else */
 685             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 686             /*                     #define INNERFLOPS INNERFLOPS+3 */
 687             /*                 #endif */
 688             /*                 ## Check for force inside potential check, i.e. this means we already did the potential part */
 689             /*                 #if 'Force' in KERNEL_VF */
 690             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq{I}{J});
 691             /*                     #define INNERFLOPS INNERFLOPS+2 */
 692             /*                 #endif */
 693             /*             #elif KERNEL_VF=='Force' */
 694             /*                 ## Force-only LennardJones makes it possible to save 1 flop (they do add up...) */
 695             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_{I}{J},rinvsix,c6_{I}{J}),_mm_mul_ps(rinvsix,rinvsq{I}{J}));
 696             /*                 #define INNERFLOPS INNERFLOPS+4 */
 697             /*             #endif */
 698
 699             /*         #elif KERNEL_VDW=='CubicSplineTable' */
 700
 701             /* CUBIC SPLINE TABLE DISPERSION */
 702             /*             #if 'Table' in KERNEL_ELEC */
 703             vfitab           = _mm_add_epi32(vfitab,ifour);
 704             /*             #endif                     */
 705             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 706             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 707             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 708             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 709             _MM_TRANSPOSE4_PS(Y,F,G,H);
 710             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 711             /*             #define INNERFLOPS INNERFLOPS+4 */
 712             /*             #if 'Potential' in KERNEL_VF */
 713             VV               = _mm_macc_ps(vfeps,Fp,Y);
 714             vvdw6            = _mm_mul_ps(c6_{I}{J},VV);
 715             /*                 #define INNERFLOPS INNERFLOPS+3 */
 716             /*             #endif */
 717             /*             #if 'Force' in KERNEL_VF */
 718             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 719             fvdw6            = _mm_mul_ps(c6_{I}{J},FF);
 720             /*                 #define INNERFLOPS INNERFLOPS+4 */
 721             /*             #endif */
 722
 723             /* CUBIC SPLINE TABLE REPULSION */
 724             vfitab           = _mm_add_epi32(vfitab,ifour);
 725             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 726             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 727             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 728             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 729             _MM_TRANSPOSE4_PS(Y,F,G,H);
 730             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 731             /*             #define INNERFLOPS INNERFLOPS+4 */
 732             /*             #if 'Potential' in KERNEL_VF */
 733             VV               = _mm_macc_ps(vfeps,Fp,Y);
 734             vvdw12           = _mm_mul_ps(c12_{I}{J},VV);
 735             /*                 #define INNERFLOPS INNERFLOPS+3 */
 736             /*             #endif */
 737             /*             #if 'Force' in KERNEL_VF */
 738             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 739             fvdw12           = _mm_mul_ps(c12_{I}{J},FF);
 740             /*                 #define INNERFLOPS INNERFLOPS+5 */
 741             /*             #endif */
 742             /*             #if 'Potential' in KERNEL_VF */
 743             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 744             /*                 #define INNERFLOPS INNERFLOPS+1 */
 745             /*             #endif */
 746             /*             #if 'Force' in KERNEL_VF */
 747             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv{I}{J})));
 748             /*                 #define INNERFLOPS INNERFLOPS+4 */
 749             /*             #endif */
 750             /*         #endif */
 751             /*         ## End of check for vdw interaction forms */
 752             /*     #endif */
 753             /*     ## END OF VDW INTERACTION CHECK FOR PAIR I-J */
 754
 755             /*     #if 'switch' in INTERACTION_FLAGS[I][J] */
 756             d                = _mm_sub_ps(r{I}{J},rswitch);
 757             d                = _mm_max_ps(d,_mm_setzero_ps());
 758             d2               = _mm_mul_ps(d,d);
 759             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 760             /*         #define INNERFLOPS INNERFLOPS+10 */
 761
 762             /*         #if 'Force' in KERNEL_VF */
 763             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 764             /*             #define INNERFLOPS INNERFLOPS+5 */
 765             /*         #endif */
 766
 767             /* Evaluate switch function */
 768             /*         #if 'Force' in KERNEL_VF */
 769             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 770             /*             #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
 771             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv{I}{J},_mm_mul_ps(velec,dsw)) );
 772             /*                 #define INNERFLOPS INNERFLOPS+4 */
 773             /*             #endif */
 774             /*             #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
 775             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv{I}{J},_mm_mul_ps(vvdw,dsw)) );
 776             /*                 #define INNERFLOPS INNERFLOPS+4 */
 777             /*             #endif */
 778             /*         #endif */
 779             /*         #if 'Potential' in KERNEL_VF */
 780             /*             #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
 781             velec            = _mm_mul_ps(velec,sw);
 782             /*                 #define INNERFLOPS INNERFLOPS+1 */
 783             /*             #endif */
 784             /*             #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
 785             vvdw             = _mm_mul_ps(vvdw,sw);
 786             /*                 #define INNERFLOPS INNERFLOPS+1 */
 787             /*             #endif */
 788             /*         #endif */
 789             /*     #endif */
 790             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 791             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 792             cutoff_mask      = _mm_cmplt_ps(rsq{I}{J},rcutoff2);
 793             /*         #define INNERFLOPS INNERFLOPS+1 */
 794             /*     #endif */
 795
 796             /*     #if 'Potential' in KERNEL_VF */
 797             /* Update potential sum for this i atom from the interaction with this j atom. */
 798             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 799             /*             #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 800             velec            = _mm_and_ps(velec,cutoff_mask);
 801             /*                 #define INNERFLOPS INNERFLOPS+1 */
 802             /*             #endif                                       */
 803             /*             #if ROUND == 'Epilogue' */
 804             velec            = _mm_andnot_ps(dummy_mask,velec);
 805             /*             #endif */
 806             velecsum         = _mm_add_ps(velecsum,velec);
 807             /*             #define INNERFLOPS INNERFLOPS+1 */
 808             /*             #if KERNEL_ELEC=='GeneralizedBorn' */
 809             /*             #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 810             vgb              = _mm_and_ps(vgb,cutoff_mask);
 811             /*                 #define INNERFLOPS INNERFLOPS+1 */
 812             /*             #endif                                       */
 813             /*             #if ROUND == 'Epilogue' */
 814             vgb              = _mm_andnot_ps(dummy_mask,vgb);
 815             /*             #endif */
 816             vgbsum           = _mm_add_ps(vgbsum,vgb);
 817             /*                 #define INNERFLOPS INNERFLOPS+1 */
 818             /*             #endif */
 819             /*         #endif */
 820             /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 821             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 822             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 823             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 824             /*                 #define INNERFLOPS INNERFLOPS+1 */
 825             /*             #endif                                       */
 826             /*             #if ROUND == 'Epilogue' */
 827             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 828             /*             #endif */
 829             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 830             /*             #define INNERFLOPS INNERFLOPS+1 */
 831             /*         #endif */
 832             /*     #endif */
 833
 834             /*     #if 'Force' in KERNEL_VF */
 835
 836             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] and 'vdw' in INTERACTION_FLAGS[I][J] */
 837             fscal            = _mm_add_ps(felec,fvdw);
 838             /*             #define INNERFLOPS INNERFLOPS+1 */
 839             /*         #elif 'electrostatics' in INTERACTION_FLAGS[I][J] */
 840             fscal            = felec;
 841             /*         #elif 'vdw' in INTERACTION_FLAGS[I][J] */
 842             fscal            = fvdw;
 843             /*        #endif */
 844
 845             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 846             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 847             fscal            = _mm_and_ps(fscal,cutoff_mask);
 848             /*                 #define INNERFLOPS INNERFLOPS+1 */
 849             /*             #endif                                       */
 850
 851             /*             #if ROUND == 'Epilogue' */
 852             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 853             /*             #endif */
 854
 855             /* ## Construction of vectorial force built into FMA instructions now */
 856             /* #define INNERFLOPS INNERFLOPS+3      */
 857
 858              /* Update vectorial force */
 859             fix{I}             = _mm_macc_ps(dx{I}{J},fscal,fix{I});
 860             fiy{I}             = _mm_macc_ps(dy{I}{J},fscal,fiy{I});
 861             fiz{I}             = _mm_macc_ps(dz{I}{J},fscal,fiz{I});
 862             /*             #define INNERFLOPS INNERFLOPS+6 */
 863
 864             /* #if GEOMETRY_I == 'Particle'             */
 865             /*     #if ROUND == 'Loop' */
 866             fjptrA             = f+j_coord_offsetA;
 867             fjptrB             = f+j_coord_offsetB;
 868             fjptrC             = f+j_coord_offsetC;
 869             fjptrD             = f+j_coord_offsetD;
 870             /*     #else */
 871             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 872             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 873             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 874             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 875             /*     #endif */
 876             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 877                                                    _mm_mul_ps(dx{I}{J},fscal),
 878                                                    _mm_mul_ps(dy{I}{J},fscal),
 879                                                    _mm_mul_ps(dz{I}{J},fscal));
 880             /*     #define INNERFLOPS INNERFLOPS+3      */
 881             /* #else                                    */
 882             fjx{J}             = _mm_macc_ps(dx{I}{J},fscal,fjx{J});
 883             fjy{J}             = _mm_macc_ps(dy{I}{J},fscal,fjy{J});
 884             fjz{J}             = _mm_macc_ps(dz{I}{J},fscal,fjz{J});
 885             /*     #define INNERFLOPS INNERFLOPS+3      */
 886             /* #endif                                   */
 887
 888             /*     #endif */
 889
 890             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 891             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 892             /*         #if 0    ## This and next two lines is a hack to maintain indentation in template file */
 893             {
 894                 /*     #endif */
 895             }
 896             /*     #endif */
 897             /*    ## End of check for the interaction being outside the cutoff */
 898
 899             /* #endfor */
 900             /* ## End of loop over i-j interaction pairs */
 901
 902             /* #if GEOMETRY_I != 'Particle' */
 903             /*     #if ROUND == 'Loop' */
 904             fjptrA             = f+j_coord_offsetA;
 905             fjptrB             = f+j_coord_offsetB;
 906             fjptrC             = f+j_coord_offsetC;
 907             fjptrD             = f+j_coord_offsetD;
 908             /*     #else */
 909             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 910             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 911             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 912             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 913             /*     #endif */
 914             /* #endif */
 915
 916             /* #if 'Water' in GEOMETRY_I and GEOMETRY_J == 'Particle' */
 917             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 918             /* #elif GEOMETRY_J == 'Water3'               */
 919             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 920                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 921             /*     #define INNERFLOPS INNERFLOPS+9      */
 922             /* #elif GEOMETRY_J == 'Water4'             */
 923             /*     #if 0 in PARTICLES_J                 */
 924             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 925                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
 926                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 927             /*     #define INNERFLOPS INNERFLOPS+12     */
 928             /*     #else                                */
 929             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
 930                                                    fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 931             /*     #define INNERFLOPS INNERFLOPS+9      */
 932             /*     #endif                               */
 933             /* #endif                                   */
 934
 935             /* Inner loop uses {INNERFLOPS} flops */
 936         }
 937
 938         /* #endfor */
 939
 940         /* End of innermost loop */
 941
 942         /* #if 'Force' in KERNEL_VF */
 943         /*     #if GEOMETRY_I == 'Particle'            */
 944         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 945                                               f+i_coord_offset,fshift+i_shift_offset);
 946         /*         #define OUTERFLOPS OUTERFLOPS+6     */
 947         /*     #elif GEOMETRY_I == 'Water3'            */
 948         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 949                                               f+i_coord_offset,fshift+i_shift_offset);
 950         /*         #define OUTERFLOPS OUTERFLOPS+18    */
 951         /*     #elif GEOMETRY_I == 'Water4'            */
 952         /*         #if 0 in PARTICLES_I                */
 953         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 954                                               f+i_coord_offset,fshift+i_shift_offset);
 955         /*             #define OUTERFLOPS OUTERFLOPS+24    */
 956         /*         #else                               */
 957         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 958                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 959         /*             #define OUTERFLOPS OUTERFLOPS+18    */
 960         /*         #endif                              */
 961         /*     #endif                                  */
 962         /* #endif                                      */
 963
 964         /* #if 'Potential' in KERNEL_VF */
 965         ggid                        = gid[iidx];
 966         /* Update potential energies */
 967         /*     #if KERNEL_ELEC != 'None' */
 968         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 969         /*         #define OUTERFLOPS OUTERFLOPS+1 */
 970         /*     #endif */
 971         /*     #if 'GeneralizedBorn' in KERNEL_ELEC */
 972         gmx_mm_update_1pot_ps(vgbsum,kernel_data->energygrp_polarization+ggid);
 973         /*         #define OUTERFLOPS OUTERFLOPS+1 */
 974         /*     #endif */
 975         /*     #if KERNEL_VDW != 'None' */
 976         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 977         /*         #define OUTERFLOPS OUTERFLOPS+1 */
 978         /*     #endif */
 979         /* #endif */
 980         /*     #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
 981         dvdasum = _mm_mul_ps(dvdasum, _mm_mul_ps(isai{I},isai{I}));
 982         gmx_mm_update_1pot_ps(dvdasum,dvda+inr);
 983         /*     #endif */
 984
 985         /* Increment number of inner iterations */
 986         inneriter                  += j_index_end - j_index_start;
 987
 988         /* Outer loop uses {OUTERFLOPS} flops */
 989     }
 990
 991     /* Increment number of outer iterations */
 992     outeriter        += nri;
 993
 994     /* Update outer/inner flops */
 995     /* ## NB: This is not important, it just affects the flopcount. However, since our preprocessor is */
 996     /* ## primitive and replaces aggressively even in strings inside these directives, we need to      */
 997     /* ## assemble the main part of the name (containing KERNEL/ELEC/VDW) directly in the source.      */
 998     /* #if GEOMETRY_I == 'Water3'            */
 999     /*     #define ISUFFIX '_W3'             */
1000     /* #elif GEOMETRY_I == 'Water4'          */
1001     /*     #define ISUFFIX '_W4'             */
1002     /* #else                                 */
1003     /*     #define ISUFFIX ''                */
1004     /* #endif                                */
1005     /* #if GEOMETRY_J == 'Water3'            */
1006     /*     #define JSUFFIX 'W3'              */
1007     /* #elif GEOMETRY_J == 'Water4'          */
1008     /*     #define JSUFFIX 'W4'              */
1009     /* #else                                 */
1010     /*     #define JSUFFIX ''                */
1011     /* #endif                                */
1012     /* #if 'PotentialAndForce' in KERNEL_VF  */
1013     /*     #define VFSUFFIX  '_VF'           */
1014     /* #elif 'Potential' in KERNEL_VF        */
1015     /*     #define VFSUFFIX '_V'             */
1016     /* #else                                 */
1017     /*     #define VFSUFFIX '_F'             */
1018     /* #endif                                */
1019
1020     /* #if KERNEL_ELEC != 'None' and KERNEL_VDW != 'None' */
1021     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1022     /* #elif KERNEL_ELEC != 'None' */
1023     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1024     /* #else */
1025     inc_nrnb(nrnb,eNR_NBKERNEL_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1026     /* #endif  */
1027 }