src/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecGB_VdwLJ_GeomP1P1_sse4_1_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_sse4_1_single.h"
  34 #include "kernelutil_x86_sse4_1_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwLJ_GeomP1P1_VF_sse4_1_single
  38  * Electrostatics interaction: GeneralizedBorn
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecGB_VdwLJ_GeomP1P1_VF_sse4_1_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  72     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  73     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  74     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  75     real             *charge;
  76     __m128i          gbitab;
  77     __m128           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
  78     __m128           minushalf = _mm_set1_ps(-0.5);
  79     real             *invsqrta,*dvda,*gbtab;
  80     int              nvdwtype;
  81     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  82     int              *vdwtype;
  83     real             *vdwparam;
  84     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  85     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  86     __m128i          vfitab;
  87     __m128i          ifour       = _mm_set1_epi32(4);
  88     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  89     real             *vftab;
  90     __m128           dummy_mask,cutoff_mask;
  91     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  92     __m128           one     = _mm_set1_ps(1.0);
  93     __m128           two     = _mm_set1_ps(2.0);
  94     x                = xx[0];
  95     f                = ff[0];
  96
  97     nri              = nlist->nri;
  98     iinr             = nlist->iinr;
  99     jindex           = nlist->jindex;
 100     jjnr             = nlist->jjnr;
 101     shiftidx         = nlist->shift;
 102     gid              = nlist->gid;
 103     shiftvec         = fr->shift_vec[0];
 104     fshift           = fr->fshift[0];
 105     facel            = _mm_set1_ps(fr->epsfac);
 106     charge           = mdatoms->chargeA;
 107     nvdwtype         = fr->ntype;
 108     vdwparam         = fr->nbfp;
 109     vdwtype          = mdatoms->typeA;
 110
 111     invsqrta         = fr->invsqrta;
 112     dvda             = fr->dvda;
 113     gbtabscale       = _mm_set1_ps(fr->gbtab.scale);
 114     gbtab            = fr->gbtab.data;
 115     gbinvepsdiff     = _mm_set1_ps((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 116
 117     /* Avoid stupid compiler warnings */
 118     jnrA = jnrB = jnrC = jnrD = 0;
 119     j_coord_offsetA = 0;
 120     j_coord_offsetB = 0;
 121     j_coord_offsetC = 0;
 122     j_coord_offsetD = 0;
 123
 124     outeriter        = 0;
 125     inneriter        = 0;
 126
 127     for(iidx=0;iidx<4*DIM;iidx++)
 128     {
 129         scratch[iidx] = 0.0;
 130     }
 131
 132     /* Start outer loop over neighborlists */
 133     for(iidx=0; iidx<nri; iidx++)
 134     {
 135         /* Load shift vector for this list */
 136         i_shift_offset   = DIM*shiftidx[iidx];
 137
 138         /* Load limits for loop over neighbors */
 139         j_index_start    = jindex[iidx];
 140         j_index_end      = jindex[iidx+1];
 141
 142         /* Get outer coordinate index */
 143         inr              = iinr[iidx];
 144         i_coord_offset   = DIM*inr;
 145
 146         /* Load i particle coords and add shift vector */
 147         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 148
 149         fix0             = _mm_setzero_ps();
 150         fiy0             = _mm_setzero_ps();
 151         fiz0             = _mm_setzero_ps();
 152
 153         /* Load parameters for i particles */
 154         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 155         isai0            = _mm_load1_ps(invsqrta+inr+0);
 156         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 157
 158         /* Reset potential sums */
 159         velecsum         = _mm_setzero_ps();
 160         vgbsum           = _mm_setzero_ps();
 161         vvdwsum          = _mm_setzero_ps();
 162         dvdasum          = _mm_setzero_ps();
 163
 164         /* Start inner kernel loop */
 165         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 166         {
 167
 168             /* Get j neighbor index, and coordinate index */
 169             jnrA             = jjnr[jidx];
 170             jnrB             = jjnr[jidx+1];
 171             jnrC             = jjnr[jidx+2];
 172             jnrD             = jjnr[jidx+3];
 173             j_coord_offsetA  = DIM*jnrA;
 174             j_coord_offsetB  = DIM*jnrB;
 175             j_coord_offsetC  = DIM*jnrC;
 176             j_coord_offsetD  = DIM*jnrD;
 177
 178             /* load j atom coordinates */
 179             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 180                                               x+j_coord_offsetC,x+j_coord_offsetD,
 181                                               &jx0,&jy0,&jz0);
 182
 183             /* Calculate displacement vector */
 184             dx00             = _mm_sub_ps(ix0,jx0);
 185             dy00             = _mm_sub_ps(iy0,jy0);
 186             dz00             = _mm_sub_ps(iz0,jz0);
 187
 188             /* Calculate squared distance and things based on it */
 189             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 190
 191             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 192
 193             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 194
 195             /* Load parameters for j particles */
 196             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 197                                                               charge+jnrC+0,charge+jnrD+0);
 198             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 199                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 200             vdwjidx0A        = 2*vdwtype[jnrA+0];
 201             vdwjidx0B        = 2*vdwtype[jnrB+0];
 202             vdwjidx0C        = 2*vdwtype[jnrC+0];
 203             vdwjidx0D        = 2*vdwtype[jnrD+0];
 204
 205             /**************************
 206              * CALCULATE INTERACTIONS *
 207              **************************/
 208
 209             r00              = _mm_mul_ps(rsq00,rinv00);
 210
 211             /* Compute parameters for interactions between i and j atoms */
 212             qq00             = _mm_mul_ps(iq0,jq0);
 213             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 214                                          vdwparam+vdwioffset0+vdwjidx0B,
 215                                          vdwparam+vdwioffset0+vdwjidx0C,
 216                                          vdwparam+vdwioffset0+vdwjidx0D,
 217                                          &c6_00,&c12_00);
 218
 219             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 220             isaprod          = _mm_mul_ps(isai0,isaj0);
 221             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 222             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 223
 224             /* Calculate generalized born table index - this is a separate table from the normal one,
 225              * but we use the same procedure by multiplying r with scale and truncating to integer.
 226              */
 227             rt               = _mm_mul_ps(r00,gbscale);
 228             gbitab           = _mm_cvttps_epi32(rt);
 229             gbeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 230             gbitab           = _mm_slli_epi32(gbitab,2);
 231             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 232             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 233             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 234             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 235             _MM_TRANSPOSE4_PS(Y,F,G,H);
 236             Heps             = _mm_mul_ps(gbeps,H);
 237             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 238             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 239             vgb              = _mm_mul_ps(gbqqfactor,VV);
 240
 241             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 242             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 243             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 244             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 245             fjptrA           = dvda+jnrA;
 246             fjptrB           = dvda+jnrB;
 247             fjptrC           = dvda+jnrC;
 248             fjptrD           = dvda+jnrD;
 249             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 250             velec            = _mm_mul_ps(qq00,rinv00);
 251             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 252
 253             /* LENNARD-JONES DISPERSION/REPULSION */
 254
 255             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 256             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 257             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 258             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 259             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 260
 261             /* Update potential sum for this i atom from the interaction with this j atom. */
 262             velecsum         = _mm_add_ps(velecsum,velec);
 263             vgbsum           = _mm_add_ps(vgbsum,vgb);
 264             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 265
 266             fscal            = _mm_add_ps(felec,fvdw);
 267
 268             /* Calculate temporary vectorial force */
 269             tx               = _mm_mul_ps(fscal,dx00);
 270             ty               = _mm_mul_ps(fscal,dy00);
 271             tz               = _mm_mul_ps(fscal,dz00);
 272
 273             /* Update vectorial force */
 274             fix0             = _mm_add_ps(fix0,tx);
 275             fiy0             = _mm_add_ps(fiy0,ty);
 276             fiz0             = _mm_add_ps(fiz0,tz);
 277
 278             fjptrA             = f+j_coord_offsetA;
 279             fjptrB             = f+j_coord_offsetB;
 280             fjptrC             = f+j_coord_offsetC;
 281             fjptrD             = f+j_coord_offsetD;
 282             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 283
 284             /* Inner loop uses 71 flops */
 285         }
 286
 287         if(jidx<j_index_end)
 288         {
 289
 290             /* Get j neighbor index, and coordinate index */
 291             jnrlistA         = jjnr[jidx];
 292             jnrlistB         = jjnr[jidx+1];
 293             jnrlistC         = jjnr[jidx+2];
 294             jnrlistD         = jjnr[jidx+3];
 295             /* Sign of each element will be negative for non-real atoms.
 296              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 297              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 298              */
 299             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 300             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 301             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 302             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 303             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 304             j_coord_offsetA  = DIM*jnrA;
 305             j_coord_offsetB  = DIM*jnrB;
 306             j_coord_offsetC  = DIM*jnrC;
 307             j_coord_offsetD  = DIM*jnrD;
 308
 309             /* load j atom coordinates */
 310             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 311                                               x+j_coord_offsetC,x+j_coord_offsetD,
 312                                               &jx0,&jy0,&jz0);
 313
 314             /* Calculate displacement vector */
 315             dx00             = _mm_sub_ps(ix0,jx0);
 316             dy00             = _mm_sub_ps(iy0,jy0);
 317             dz00             = _mm_sub_ps(iz0,jz0);
 318
 319             /* Calculate squared distance and things based on it */
 320             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 321
 322             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 323
 324             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 325
 326             /* Load parameters for j particles */
 327             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 328                                                               charge+jnrC+0,charge+jnrD+0);
 329             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 330                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 331             vdwjidx0A        = 2*vdwtype[jnrA+0];
 332             vdwjidx0B        = 2*vdwtype[jnrB+0];
 333             vdwjidx0C        = 2*vdwtype[jnrC+0];
 334             vdwjidx0D        = 2*vdwtype[jnrD+0];
 335
 336             /**************************
 337              * CALCULATE INTERACTIONS *
 338              **************************/
 339
 340             r00              = _mm_mul_ps(rsq00,rinv00);
 341             r00              = _mm_andnot_ps(dummy_mask,r00);
 342
 343             /* Compute parameters for interactions between i and j atoms */
 344             qq00             = _mm_mul_ps(iq0,jq0);
 345             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 346                                          vdwparam+vdwioffset0+vdwjidx0B,
 347                                          vdwparam+vdwioffset0+vdwjidx0C,
 348                                          vdwparam+vdwioffset0+vdwjidx0D,
 349                                          &c6_00,&c12_00);
 350
 351             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 352             isaprod          = _mm_mul_ps(isai0,isaj0);
 353             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 354             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 355
 356             /* Calculate generalized born table index - this is a separate table from the normal one,
 357              * but we use the same procedure by multiplying r with scale and truncating to integer.
 358              */
 359             rt               = _mm_mul_ps(r00,gbscale);
 360             gbitab           = _mm_cvttps_epi32(rt);
 361             gbeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 362             gbitab           = _mm_slli_epi32(gbitab,2);
 363             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 364             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 365             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 366             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 367             _MM_TRANSPOSE4_PS(Y,F,G,H);
 368             Heps             = _mm_mul_ps(gbeps,H);
 369             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 370             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 371             vgb              = _mm_mul_ps(gbqqfactor,VV);
 372
 373             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 374             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 375             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 376             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 377             /* 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. */
 378             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 379             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 380             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 381             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 382             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 383             velec            = _mm_mul_ps(qq00,rinv00);
 384             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 385
 386             /* LENNARD-JONES DISPERSION/REPULSION */
 387
 388             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 389             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 390             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 391             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 392             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 393
 394             /* Update potential sum for this i atom from the interaction with this j atom. */
 395             velec            = _mm_andnot_ps(dummy_mask,velec);
 396             velecsum         = _mm_add_ps(velecsum,velec);
 397             vgb              = _mm_andnot_ps(dummy_mask,vgb);
 398             vgbsum           = _mm_add_ps(vgbsum,vgb);
 399             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 400             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 401
 402             fscal            = _mm_add_ps(felec,fvdw);
 403
 404             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 405
 406             /* Calculate temporary vectorial force */
 407             tx               = _mm_mul_ps(fscal,dx00);
 408             ty               = _mm_mul_ps(fscal,dy00);
 409             tz               = _mm_mul_ps(fscal,dz00);
 410
 411             /* Update vectorial force */
 412             fix0             = _mm_add_ps(fix0,tx);
 413             fiy0             = _mm_add_ps(fiy0,ty);
 414             fiz0             = _mm_add_ps(fiz0,tz);
 415
 416             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 417             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 418             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 419             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 420             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 421
 422             /* Inner loop uses 72 flops */
 423         }
 424
 425         /* End of innermost loop */
 426
 427         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 428                                               f+i_coord_offset,fshift+i_shift_offset);
 429
 430         ggid                        = gid[iidx];
 431         /* Update potential energies */
 432         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 433         gmx_mm_update_1pot_ps(vgbsum,kernel_data->energygrp_polarization+ggid);
 434         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 435         dvdasum = _mm_mul_ps(dvdasum, _mm_mul_ps(isai0,isai0));
 436         gmx_mm_update_1pot_ps(dvdasum,dvda+inr);
 437
 438         /* Increment number of inner iterations */
 439         inneriter                  += j_index_end - j_index_start;
 440
 441         /* Outer loop uses 10 flops */
 442     }
 443
 444     /* Increment number of outer iterations */
 445     outeriter        += nri;
 446
 447     /* Update outer/inner flops */
 448
 449     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*10 + inneriter*72);
 450 }
 451 /*
 452  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwLJ_GeomP1P1_F_sse4_1_single
 453  * Electrostatics interaction: GeneralizedBorn
 454  * VdW interaction:            LennardJones
 455  * Geometry:                   Particle-Particle
 456  * Calculate force/pot:        Force
 457  */
 458 void
 459 nb_kernel_ElecGB_VdwLJ_GeomP1P1_F_sse4_1_single
 460                     (t_nblist * gmx_restrict                nlist,
 461                      rvec * gmx_restrict                    xx,
 462                      rvec * gmx_restrict                    ff,
 463                      t_forcerec * gmx_restrict              fr,
 464                      t_mdatoms * gmx_restrict               mdatoms,
 465                      nb_kernel_data_t * gmx_restrict        kernel_data,
 466                      t_nrnb * gmx_restrict                  nrnb)
 467 {
 468     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 469      * just 0 for non-waters.
 470      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 471      * jnr indices corresponding to data put in the four positions in the SIMD register.
 472      */
 473     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 474     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 475     int              jnrA,jnrB,jnrC,jnrD;
 476     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 477     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 478     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 479     real             rcutoff_scalar;
 480     real             *shiftvec,*fshift,*x,*f;
 481     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 482     real             scratch[4*DIM];
 483     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 484     int              vdwioffset0;
 485     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 486     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 487     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 488     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 489     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 490     real             *charge;
 491     __m128i          gbitab;
 492     __m128           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
 493     __m128           minushalf = _mm_set1_ps(-0.5);
 494     real             *invsqrta,*dvda,*gbtab;
 495     int              nvdwtype;
 496     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 497     int              *vdwtype;
 498     real             *vdwparam;
 499     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 500     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 501     __m128i          vfitab;
 502     __m128i          ifour       = _mm_set1_epi32(4);
 503     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 504     real             *vftab;
 505     __m128           dummy_mask,cutoff_mask;
 506     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 507     __m128           one     = _mm_set1_ps(1.0);
 508     __m128           two     = _mm_set1_ps(2.0);
 509     x                = xx[0];
 510     f                = ff[0];
 511
 512     nri              = nlist->nri;
 513     iinr             = nlist->iinr;
 514     jindex           = nlist->jindex;
 515     jjnr             = nlist->jjnr;
 516     shiftidx         = nlist->shift;
 517     gid              = nlist->gid;
 518     shiftvec         = fr->shift_vec[0];
 519     fshift           = fr->fshift[0];
 520     facel            = _mm_set1_ps(fr->epsfac);
 521     charge           = mdatoms->chargeA;
 522     nvdwtype         = fr->ntype;
 523     vdwparam         = fr->nbfp;
 524     vdwtype          = mdatoms->typeA;
 525
 526     invsqrta         = fr->invsqrta;
 527     dvda             = fr->dvda;
 528     gbtabscale       = _mm_set1_ps(fr->gbtab.scale);
 529     gbtab            = fr->gbtab.data;
 530     gbinvepsdiff     = _mm_set1_ps((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 531
 532     /* Avoid stupid compiler warnings */
 533     jnrA = jnrB = jnrC = jnrD = 0;
 534     j_coord_offsetA = 0;
 535     j_coord_offsetB = 0;
 536     j_coord_offsetC = 0;
 537     j_coord_offsetD = 0;
 538
 539     outeriter        = 0;
 540     inneriter        = 0;
 541
 542     for(iidx=0;iidx<4*DIM;iidx++)
 543     {
 544         scratch[iidx] = 0.0;
 545     }
 546
 547     /* Start outer loop over neighborlists */
 548     for(iidx=0; iidx<nri; iidx++)
 549     {
 550         /* Load shift vector for this list */
 551         i_shift_offset   = DIM*shiftidx[iidx];
 552
 553         /* Load limits for loop over neighbors */
 554         j_index_start    = jindex[iidx];
 555         j_index_end      = jindex[iidx+1];
 556
 557         /* Get outer coordinate index */
 558         inr              = iinr[iidx];
 559         i_coord_offset   = DIM*inr;
 560
 561         /* Load i particle coords and add shift vector */
 562         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 563
 564         fix0             = _mm_setzero_ps();
 565         fiy0             = _mm_setzero_ps();
 566         fiz0             = _mm_setzero_ps();
 567
 568         /* Load parameters for i particles */
 569         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 570         isai0            = _mm_load1_ps(invsqrta+inr+0);
 571         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 572
 573         dvdasum          = _mm_setzero_ps();
 574
 575         /* Start inner kernel loop */
 576         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 577         {
 578
 579             /* Get j neighbor index, and coordinate index */
 580             jnrA             = jjnr[jidx];
 581             jnrB             = jjnr[jidx+1];
 582             jnrC             = jjnr[jidx+2];
 583             jnrD             = jjnr[jidx+3];
 584             j_coord_offsetA  = DIM*jnrA;
 585             j_coord_offsetB  = DIM*jnrB;
 586             j_coord_offsetC  = DIM*jnrC;
 587             j_coord_offsetD  = DIM*jnrD;
 588
 589             /* load j atom coordinates */
 590             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 591                                               x+j_coord_offsetC,x+j_coord_offsetD,
 592                                               &jx0,&jy0,&jz0);
 593
 594             /* Calculate displacement vector */
 595             dx00             = _mm_sub_ps(ix0,jx0);
 596             dy00             = _mm_sub_ps(iy0,jy0);
 597             dz00             = _mm_sub_ps(iz0,jz0);
 598
 599             /* Calculate squared distance and things based on it */
 600             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 601
 602             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 603
 604             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 605
 606             /* Load parameters for j particles */
 607             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 608                                                               charge+jnrC+0,charge+jnrD+0);
 609             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 610                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 611             vdwjidx0A        = 2*vdwtype[jnrA+0];
 612             vdwjidx0B        = 2*vdwtype[jnrB+0];
 613             vdwjidx0C        = 2*vdwtype[jnrC+0];
 614             vdwjidx0D        = 2*vdwtype[jnrD+0];
 615
 616             /**************************
 617              * CALCULATE INTERACTIONS *
 618              **************************/
 619
 620             r00              = _mm_mul_ps(rsq00,rinv00);
 621
 622             /* Compute parameters for interactions between i and j atoms */
 623             qq00             = _mm_mul_ps(iq0,jq0);
 624             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 625                                          vdwparam+vdwioffset0+vdwjidx0B,
 626                                          vdwparam+vdwioffset0+vdwjidx0C,
 627                                          vdwparam+vdwioffset0+vdwjidx0D,
 628                                          &c6_00,&c12_00);
 629
 630             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 631             isaprod          = _mm_mul_ps(isai0,isaj0);
 632             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 633             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 634
 635             /* Calculate generalized born table index - this is a separate table from the normal one,
 636              * but we use the same procedure by multiplying r with scale and truncating to integer.
 637              */
 638             rt               = _mm_mul_ps(r00,gbscale);
 639             gbitab           = _mm_cvttps_epi32(rt);
 640             gbeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 641             gbitab           = _mm_slli_epi32(gbitab,2);
 642             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 643             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 644             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 645             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 646             _MM_TRANSPOSE4_PS(Y,F,G,H);
 647             Heps             = _mm_mul_ps(gbeps,H);
 648             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 649             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 650             vgb              = _mm_mul_ps(gbqqfactor,VV);
 651
 652             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 653             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 654             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 655             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 656             fjptrA           = dvda+jnrA;
 657             fjptrB           = dvda+jnrB;
 658             fjptrC           = dvda+jnrC;
 659             fjptrD           = dvda+jnrD;
 660             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 661             velec            = _mm_mul_ps(qq00,rinv00);
 662             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 663
 664             /* LENNARD-JONES DISPERSION/REPULSION */
 665
 666             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 667             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 668
 669             fscal            = _mm_add_ps(felec,fvdw);
 670
 671             /* Calculate temporary vectorial force */
 672             tx               = _mm_mul_ps(fscal,dx00);
 673             ty               = _mm_mul_ps(fscal,dy00);
 674             tz               = _mm_mul_ps(fscal,dz00);
 675
 676             /* Update vectorial force */
 677             fix0             = _mm_add_ps(fix0,tx);
 678             fiy0             = _mm_add_ps(fiy0,ty);
 679             fiz0             = _mm_add_ps(fiz0,tz);
 680
 681             fjptrA             = f+j_coord_offsetA;
 682             fjptrB             = f+j_coord_offsetB;
 683             fjptrC             = f+j_coord_offsetC;
 684             fjptrD             = f+j_coord_offsetD;
 685             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 686
 687             /* Inner loop uses 64 flops */
 688         }
 689
 690         if(jidx<j_index_end)
 691         {
 692
 693             /* Get j neighbor index, and coordinate index */
 694             jnrlistA         = jjnr[jidx];
 695             jnrlistB         = jjnr[jidx+1];
 696             jnrlistC         = jjnr[jidx+2];
 697             jnrlistD         = jjnr[jidx+3];
 698             /* Sign of each element will be negative for non-real atoms.
 699              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 700              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 701              */
 702             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 703             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 704             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 705             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 706             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 707             j_coord_offsetA  = DIM*jnrA;
 708             j_coord_offsetB  = DIM*jnrB;
 709             j_coord_offsetC  = DIM*jnrC;
 710             j_coord_offsetD  = DIM*jnrD;
 711
 712             /* load j atom coordinates */
 713             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 714                                               x+j_coord_offsetC,x+j_coord_offsetD,
 715                                               &jx0,&jy0,&jz0);
 716
 717             /* Calculate displacement vector */
 718             dx00             = _mm_sub_ps(ix0,jx0);
 719             dy00             = _mm_sub_ps(iy0,jy0);
 720             dz00             = _mm_sub_ps(iz0,jz0);
 721
 722             /* Calculate squared distance and things based on it */
 723             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 724
 725             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 726
 727             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 728
 729             /* Load parameters for j particles */
 730             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 731                                                               charge+jnrC+0,charge+jnrD+0);
 732             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 733                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 734             vdwjidx0A        = 2*vdwtype[jnrA+0];
 735             vdwjidx0B        = 2*vdwtype[jnrB+0];
 736             vdwjidx0C        = 2*vdwtype[jnrC+0];
 737             vdwjidx0D        = 2*vdwtype[jnrD+0];
 738
 739             /**************************
 740              * CALCULATE INTERACTIONS *
 741              **************************/
 742
 743             r00              = _mm_mul_ps(rsq00,rinv00);
 744             r00              = _mm_andnot_ps(dummy_mask,r00);
 745
 746             /* Compute parameters for interactions between i and j atoms */
 747             qq00             = _mm_mul_ps(iq0,jq0);
 748             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 749                                          vdwparam+vdwioffset0+vdwjidx0B,
 750                                          vdwparam+vdwioffset0+vdwjidx0C,
 751                                          vdwparam+vdwioffset0+vdwjidx0D,
 752                                          &c6_00,&c12_00);
 753
 754             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 755             isaprod          = _mm_mul_ps(isai0,isaj0);
 756             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 757             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 758
 759             /* Calculate generalized born table index - this is a separate table from the normal one,
 760              * but we use the same procedure by multiplying r with scale and truncating to integer.
 761              */
 762             rt               = _mm_mul_ps(r00,gbscale);
 763             gbitab           = _mm_cvttps_epi32(rt);
 764             gbeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 765             gbitab           = _mm_slli_epi32(gbitab,2);
 766             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 767             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 768             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 769             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 770             _MM_TRANSPOSE4_PS(Y,F,G,H);
 771             Heps             = _mm_mul_ps(gbeps,H);
 772             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 773             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 774             vgb              = _mm_mul_ps(gbqqfactor,VV);
 775
 776             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 777             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 778             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 779             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 780             /* 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. */
 781             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 782             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 783             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 784             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 785             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 786             velec            = _mm_mul_ps(qq00,rinv00);
 787             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 788
 789             /* LENNARD-JONES DISPERSION/REPULSION */
 790
 791             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 792             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 793
 794             fscal            = _mm_add_ps(felec,fvdw);
 795
 796             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 797
 798             /* Calculate temporary vectorial force */
 799             tx               = _mm_mul_ps(fscal,dx00);
 800             ty               = _mm_mul_ps(fscal,dy00);
 801             tz               = _mm_mul_ps(fscal,dz00);
 802
 803             /* Update vectorial force */
 804             fix0             = _mm_add_ps(fix0,tx);
 805             fiy0             = _mm_add_ps(fiy0,ty);
 806             fiz0             = _mm_add_ps(fiz0,tz);
 807
 808             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 809             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 810             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 811             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 812             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 813
 814             /* Inner loop uses 65 flops */
 815         }
 816
 817         /* End of innermost loop */
 818
 819         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 820                                               f+i_coord_offset,fshift+i_shift_offset);
 821
 822         dvdasum = _mm_mul_ps(dvdasum, _mm_mul_ps(isai0,isai0));
 823         gmx_mm_update_1pot_ps(dvdasum,dvda+inr);
 824
 825         /* Increment number of inner iterations */
 826         inneriter                  += j_index_end - j_index_start;
 827
 828         /* Outer loop uses 7 flops */
 829     }
 830
 831     /* Increment number of outer iterations */
 832     outeriter        += nri;
 833
 834     /* Update outer/inner flops */
 835
 836     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*65);
 837 }