src/gromacs/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             dvdatmp          = _mm_andnot_ps(dummy_mask,dvdatmp);
 377             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 378             /* 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. */
 379             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 380             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 381             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 382             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 383             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 384             velec            = _mm_mul_ps(qq00,rinv00);
 385             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 386
 387             /* LENNARD-JONES DISPERSION/REPULSION */
 388
 389             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 390             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 391             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 392             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 393             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 394
 395             /* Update potential sum for this i atom from the interaction with this j atom. */
 396             velec            = _mm_andnot_ps(dummy_mask,velec);
 397             velecsum         = _mm_add_ps(velecsum,velec);
 398             vgb              = _mm_andnot_ps(dummy_mask,vgb);
 399             vgbsum           = _mm_add_ps(vgbsum,vgb);
 400             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 401             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 402
 403             fscal            = _mm_add_ps(felec,fvdw);
 404
 405             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 406
 407             /* Calculate temporary vectorial force */
 408             tx               = _mm_mul_ps(fscal,dx00);
 409             ty               = _mm_mul_ps(fscal,dy00);
 410             tz               = _mm_mul_ps(fscal,dz00);
 411
 412             /* Update vectorial force */
 413             fix0             = _mm_add_ps(fix0,tx);
 414             fiy0             = _mm_add_ps(fiy0,ty);
 415             fiz0             = _mm_add_ps(fiz0,tz);
 416
 417             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 418             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 419             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 420             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 421             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 422
 423             /* Inner loop uses 72 flops */
 424         }
 425
 426         /* End of innermost loop */
 427
 428         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 429                                               f+i_coord_offset,fshift+i_shift_offset);
 430
 431         ggid                        = gid[iidx];
 432         /* Update potential energies */
 433         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 434         gmx_mm_update_1pot_ps(vgbsum,kernel_data->energygrp_polarization+ggid);
 435         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 436         dvdasum = _mm_mul_ps(dvdasum, _mm_mul_ps(isai0,isai0));
 437         gmx_mm_update_1pot_ps(dvdasum,dvda+inr);
 438
 439         /* Increment number of inner iterations */
 440         inneriter                  += j_index_end - j_index_start;
 441
 442         /* Outer loop uses 10 flops */
 443     }
 444
 445     /* Increment number of outer iterations */
 446     outeriter        += nri;
 447
 448     /* Update outer/inner flops */
 449
 450     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*10 + inneriter*72);
 451 }
 452 /*
 453  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwLJ_GeomP1P1_F_sse4_1_single
 454  * Electrostatics interaction: GeneralizedBorn
 455  * VdW interaction:            LennardJones
 456  * Geometry:                   Particle-Particle
 457  * Calculate force/pot:        Force
 458  */
 459 void
 460 nb_kernel_ElecGB_VdwLJ_GeomP1P1_F_sse4_1_single
 461                     (t_nblist * gmx_restrict                nlist,
 462                      rvec * gmx_restrict                    xx,
 463                      rvec * gmx_restrict                    ff,
 464                      t_forcerec * gmx_restrict              fr,
 465                      t_mdatoms * gmx_restrict               mdatoms,
 466                      nb_kernel_data_t * gmx_restrict        kernel_data,
 467                      t_nrnb * gmx_restrict                  nrnb)
 468 {
 469     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 470      * just 0 for non-waters.
 471      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 472      * jnr indices corresponding to data put in the four positions in the SIMD register.
 473      */
 474     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 475     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 476     int              jnrA,jnrB,jnrC,jnrD;
 477     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 478     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 479     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 480     real             rcutoff_scalar;
 481     real             *shiftvec,*fshift,*x,*f;
 482     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 483     real             scratch[4*DIM];
 484     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 485     int              vdwioffset0;
 486     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 487     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 488     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 489     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 490     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 491     real             *charge;
 492     __m128i          gbitab;
 493     __m128           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
 494     __m128           minushalf = _mm_set1_ps(-0.5);
 495     real             *invsqrta,*dvda,*gbtab;
 496     int              nvdwtype;
 497     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 498     int              *vdwtype;
 499     real             *vdwparam;
 500     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 501     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 502     __m128i          vfitab;
 503     __m128i          ifour       = _mm_set1_epi32(4);
 504     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 505     real             *vftab;
 506     __m128           dummy_mask,cutoff_mask;
 507     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 508     __m128           one     = _mm_set1_ps(1.0);
 509     __m128           two     = _mm_set1_ps(2.0);
 510     x                = xx[0];
 511     f                = ff[0];
 512
 513     nri              = nlist->nri;
 514     iinr             = nlist->iinr;
 515     jindex           = nlist->jindex;
 516     jjnr             = nlist->jjnr;
 517     shiftidx         = nlist->shift;
 518     gid              = nlist->gid;
 519     shiftvec         = fr->shift_vec[0];
 520     fshift           = fr->fshift[0];
 521     facel            = _mm_set1_ps(fr->epsfac);
 522     charge           = mdatoms->chargeA;
 523     nvdwtype         = fr->ntype;
 524     vdwparam         = fr->nbfp;
 525     vdwtype          = mdatoms->typeA;
 526
 527     invsqrta         = fr->invsqrta;
 528     dvda             = fr->dvda;
 529     gbtabscale       = _mm_set1_ps(fr->gbtab.scale);
 530     gbtab            = fr->gbtab.data;
 531     gbinvepsdiff     = _mm_set1_ps((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 532
 533     /* Avoid stupid compiler warnings */
 534     jnrA = jnrB = jnrC = jnrD = 0;
 535     j_coord_offsetA = 0;
 536     j_coord_offsetB = 0;
 537     j_coord_offsetC = 0;
 538     j_coord_offsetD = 0;
 539
 540     outeriter        = 0;
 541     inneriter        = 0;
 542
 543     for(iidx=0;iidx<4*DIM;iidx++)
 544     {
 545         scratch[iidx] = 0.0;
 546     }
 547
 548     /* Start outer loop over neighborlists */
 549     for(iidx=0; iidx<nri; iidx++)
 550     {
 551         /* Load shift vector for this list */
 552         i_shift_offset   = DIM*shiftidx[iidx];
 553
 554         /* Load limits for loop over neighbors */
 555         j_index_start    = jindex[iidx];
 556         j_index_end      = jindex[iidx+1];
 557
 558         /* Get outer coordinate index */
 559         inr              = iinr[iidx];
 560         i_coord_offset   = DIM*inr;
 561
 562         /* Load i particle coords and add shift vector */
 563         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 564
 565         fix0             = _mm_setzero_ps();
 566         fiy0             = _mm_setzero_ps();
 567         fiz0             = _mm_setzero_ps();
 568
 569         /* Load parameters for i particles */
 570         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 571         isai0            = _mm_load1_ps(invsqrta+inr+0);
 572         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 573
 574         dvdasum          = _mm_setzero_ps();
 575
 576         /* Start inner kernel loop */
 577         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 578         {
 579
 580             /* Get j neighbor index, and coordinate index */
 581             jnrA             = jjnr[jidx];
 582             jnrB             = jjnr[jidx+1];
 583             jnrC             = jjnr[jidx+2];
 584             jnrD             = jjnr[jidx+3];
 585             j_coord_offsetA  = DIM*jnrA;
 586             j_coord_offsetB  = DIM*jnrB;
 587             j_coord_offsetC  = DIM*jnrC;
 588             j_coord_offsetD  = DIM*jnrD;
 589
 590             /* load j atom coordinates */
 591             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 592                                               x+j_coord_offsetC,x+j_coord_offsetD,
 593                                               &jx0,&jy0,&jz0);
 594
 595             /* Calculate displacement vector */
 596             dx00             = _mm_sub_ps(ix0,jx0);
 597             dy00             = _mm_sub_ps(iy0,jy0);
 598             dz00             = _mm_sub_ps(iz0,jz0);
 599
 600             /* Calculate squared distance and things based on it */
 601             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 602
 603             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 604
 605             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 606
 607             /* Load parameters for j particles */
 608             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 609                                                               charge+jnrC+0,charge+jnrD+0);
 610             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 611                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 612             vdwjidx0A        = 2*vdwtype[jnrA+0];
 613             vdwjidx0B        = 2*vdwtype[jnrB+0];
 614             vdwjidx0C        = 2*vdwtype[jnrC+0];
 615             vdwjidx0D        = 2*vdwtype[jnrD+0];
 616
 617             /**************************
 618              * CALCULATE INTERACTIONS *
 619              **************************/
 620
 621             r00              = _mm_mul_ps(rsq00,rinv00);
 622
 623             /* Compute parameters for interactions between i and j atoms */
 624             qq00             = _mm_mul_ps(iq0,jq0);
 625             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 626                                          vdwparam+vdwioffset0+vdwjidx0B,
 627                                          vdwparam+vdwioffset0+vdwjidx0C,
 628                                          vdwparam+vdwioffset0+vdwjidx0D,
 629                                          &c6_00,&c12_00);
 630
 631             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 632             isaprod          = _mm_mul_ps(isai0,isaj0);
 633             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 634             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 635
 636             /* Calculate generalized born table index - this is a separate table from the normal one,
 637              * but we use the same procedure by multiplying r with scale and truncating to integer.
 638              */
 639             rt               = _mm_mul_ps(r00,gbscale);
 640             gbitab           = _mm_cvttps_epi32(rt);
 641             gbeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 642             gbitab           = _mm_slli_epi32(gbitab,2);
 643             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 644             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 645             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 646             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 647             _MM_TRANSPOSE4_PS(Y,F,G,H);
 648             Heps             = _mm_mul_ps(gbeps,H);
 649             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 650             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 651             vgb              = _mm_mul_ps(gbqqfactor,VV);
 652
 653             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 654             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 655             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 656             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 657             fjptrA           = dvda+jnrA;
 658             fjptrB           = dvda+jnrB;
 659             fjptrC           = dvda+jnrC;
 660             fjptrD           = dvda+jnrD;
 661             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 662             velec            = _mm_mul_ps(qq00,rinv00);
 663             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 664
 665             /* LENNARD-JONES DISPERSION/REPULSION */
 666
 667             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 668             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 669
 670             fscal            = _mm_add_ps(felec,fvdw);
 671
 672             /* Calculate temporary vectorial force */
 673             tx               = _mm_mul_ps(fscal,dx00);
 674             ty               = _mm_mul_ps(fscal,dy00);
 675             tz               = _mm_mul_ps(fscal,dz00);
 676
 677             /* Update vectorial force */
 678             fix0             = _mm_add_ps(fix0,tx);
 679             fiy0             = _mm_add_ps(fiy0,ty);
 680             fiz0             = _mm_add_ps(fiz0,tz);
 681
 682             fjptrA             = f+j_coord_offsetA;
 683             fjptrB             = f+j_coord_offsetB;
 684             fjptrC             = f+j_coord_offsetC;
 685             fjptrD             = f+j_coord_offsetD;
 686             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 687
 688             /* Inner loop uses 64 flops */
 689         }
 690
 691         if(jidx<j_index_end)
 692         {
 693
 694             /* Get j neighbor index, and coordinate index */
 695             jnrlistA         = jjnr[jidx];
 696             jnrlistB         = jjnr[jidx+1];
 697             jnrlistC         = jjnr[jidx+2];
 698             jnrlistD         = jjnr[jidx+3];
 699             /* Sign of each element will be negative for non-real atoms.
 700              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 701              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 702              */
 703             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 704             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 705             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 706             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 707             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 708             j_coord_offsetA  = DIM*jnrA;
 709             j_coord_offsetB  = DIM*jnrB;
 710             j_coord_offsetC  = DIM*jnrC;
 711             j_coord_offsetD  = DIM*jnrD;
 712
 713             /* load j atom coordinates */
 714             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 715                                               x+j_coord_offsetC,x+j_coord_offsetD,
 716                                               &jx0,&jy0,&jz0);
 717
 718             /* Calculate displacement vector */
 719             dx00             = _mm_sub_ps(ix0,jx0);
 720             dy00             = _mm_sub_ps(iy0,jy0);
 721             dz00             = _mm_sub_ps(iz0,jz0);
 722
 723             /* Calculate squared distance and things based on it */
 724             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 725
 726             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 727
 728             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 729
 730             /* Load parameters for j particles */
 731             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 732                                                               charge+jnrC+0,charge+jnrD+0);
 733             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 734                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 735             vdwjidx0A        = 2*vdwtype[jnrA+0];
 736             vdwjidx0B        = 2*vdwtype[jnrB+0];
 737             vdwjidx0C        = 2*vdwtype[jnrC+0];
 738             vdwjidx0D        = 2*vdwtype[jnrD+0];
 739
 740             /**************************
 741              * CALCULATE INTERACTIONS *
 742              **************************/
 743
 744             r00              = _mm_mul_ps(rsq00,rinv00);
 745             r00              = _mm_andnot_ps(dummy_mask,r00);
 746
 747             /* Compute parameters for interactions between i and j atoms */
 748             qq00             = _mm_mul_ps(iq0,jq0);
 749             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 750                                          vdwparam+vdwioffset0+vdwjidx0B,
 751                                          vdwparam+vdwioffset0+vdwjidx0C,
 752                                          vdwparam+vdwioffset0+vdwjidx0D,
 753                                          &c6_00,&c12_00);
 754
 755             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 756             isaprod          = _mm_mul_ps(isai0,isaj0);
 757             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 758             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 759
 760             /* Calculate generalized born table index - this is a separate table from the normal one,
 761              * but we use the same procedure by multiplying r with scale and truncating to integer.
 762              */
 763             rt               = _mm_mul_ps(r00,gbscale);
 764             gbitab           = _mm_cvttps_epi32(rt);
 765             gbeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 766             gbitab           = _mm_slli_epi32(gbitab,2);
 767             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 768             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 769             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 770             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 771             _MM_TRANSPOSE4_PS(Y,F,G,H);
 772             Heps             = _mm_mul_ps(gbeps,H);
 773             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 774             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 775             vgb              = _mm_mul_ps(gbqqfactor,VV);
 776
 777             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 778             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 779             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 780             dvdatmp          = _mm_andnot_ps(dummy_mask,dvdatmp);
 781             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 782             /* 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. */
 783             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 784             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 785             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 786             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 787             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 788             velec            = _mm_mul_ps(qq00,rinv00);
 789             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 790
 791             /* LENNARD-JONES DISPERSION/REPULSION */
 792
 793             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 794             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 795
 796             fscal            = _mm_add_ps(felec,fvdw);
 797
 798             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 799
 800             /* Calculate temporary vectorial force */
 801             tx               = _mm_mul_ps(fscal,dx00);
 802             ty               = _mm_mul_ps(fscal,dy00);
 803             tz               = _mm_mul_ps(fscal,dz00);
 804
 805             /* Update vectorial force */
 806             fix0             = _mm_add_ps(fix0,tx);
 807             fiy0             = _mm_add_ps(fiy0,ty);
 808             fiz0             = _mm_add_ps(fiz0,tz);
 809
 810             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 811             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 812             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 813             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 814             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 815
 816             /* Inner loop uses 65 flops */
 817         }
 818
 819         /* End of innermost loop */
 820
 821         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 822                                               f+i_coord_offset,fshift+i_shift_offset);
 823
 824         dvdasum = _mm_mul_ps(dvdasum, _mm_mul_ps(isai0,isai0));
 825         gmx_mm_update_1pot_ps(dvdasum,dvda+inr);
 826
 827         /* Increment number of inner iterations */
 828         inneriter                  += j_index_end - j_index_start;
 829
 830         /* Outer loop uses 7 flops */
 831     }
 832
 833     /* Increment number of outer iterations */
 834     outeriter        += nri;
 835
 836     /* Update outer/inner flops */
 837
 838     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*65);
 839 }