src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecGB_VdwLJ_GeomP1P1_sse2_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_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_sse2_single.h"
  34 #include "kernelutil_x86_sse2_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwLJ_GeomP1P1_VF_sse2_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_sse2_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_cvtepi32_ps(gbitab));
 230             gbitab           = _mm_slli_epi32(gbitab,2);
 231
 232             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 233             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 234             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 235             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 236             _MM_TRANSPOSE4_PS(Y,F,G,H);
 237             Heps             = _mm_mul_ps(gbeps,H);
 238             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 239             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 240             vgb              = _mm_mul_ps(gbqqfactor,VV);
 241
 242             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 243             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 244             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 245             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 246             fjptrA           = dvda+jnrA;
 247             fjptrB           = dvda+jnrB;
 248             fjptrC           = dvda+jnrC;
 249             fjptrD           = dvda+jnrD;
 250             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 251             velec            = _mm_mul_ps(qq00,rinv00);
 252             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 253
 254             /* LENNARD-JONES DISPERSION/REPULSION */
 255
 256             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 257             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 258             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 259             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 260             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 261
 262             /* Update potential sum for this i atom from the interaction with this j atom. */
 263             velecsum         = _mm_add_ps(velecsum,velec);
 264             vgbsum           = _mm_add_ps(vgbsum,vgb);
 265             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 266
 267             fscal            = _mm_add_ps(felec,fvdw);
 268
 269             /* Calculate temporary vectorial force */
 270             tx               = _mm_mul_ps(fscal,dx00);
 271             ty               = _mm_mul_ps(fscal,dy00);
 272             tz               = _mm_mul_ps(fscal,dz00);
 273
 274             /* Update vectorial force */
 275             fix0             = _mm_add_ps(fix0,tx);
 276             fiy0             = _mm_add_ps(fiy0,ty);
 277             fiz0             = _mm_add_ps(fiz0,tz);
 278
 279             fjptrA             = f+j_coord_offsetA;
 280             fjptrB             = f+j_coord_offsetB;
 281             fjptrC             = f+j_coord_offsetC;
 282             fjptrD             = f+j_coord_offsetD;
 283             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 284
 285             /* Inner loop uses 71 flops */
 286         }
 287
 288         if(jidx<j_index_end)
 289         {
 290
 291             /* Get j neighbor index, and coordinate index */
 292             jnrlistA         = jjnr[jidx];
 293             jnrlistB         = jjnr[jidx+1];
 294             jnrlistC         = jjnr[jidx+2];
 295             jnrlistD         = jjnr[jidx+3];
 296             /* Sign of each element will be negative for non-real atoms.
 297              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 298              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 299              */
 300             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 301             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 302             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 303             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 304             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 305             j_coord_offsetA  = DIM*jnrA;
 306             j_coord_offsetB  = DIM*jnrB;
 307             j_coord_offsetC  = DIM*jnrC;
 308             j_coord_offsetD  = DIM*jnrD;
 309
 310             /* load j atom coordinates */
 311             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 312                                               x+j_coord_offsetC,x+j_coord_offsetD,
 313                                               &jx0,&jy0,&jz0);
 314
 315             /* Calculate displacement vector */
 316             dx00             = _mm_sub_ps(ix0,jx0);
 317             dy00             = _mm_sub_ps(iy0,jy0);
 318             dz00             = _mm_sub_ps(iz0,jz0);
 319
 320             /* Calculate squared distance and things based on it */
 321             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 322
 323             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 324
 325             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 326
 327             /* Load parameters for j particles */
 328             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 329                                                               charge+jnrC+0,charge+jnrD+0);
 330             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 331                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 332             vdwjidx0A        = 2*vdwtype[jnrA+0];
 333             vdwjidx0B        = 2*vdwtype[jnrB+0];
 334             vdwjidx0C        = 2*vdwtype[jnrC+0];
 335             vdwjidx0D        = 2*vdwtype[jnrD+0];
 336
 337             /**************************
 338              * CALCULATE INTERACTIONS *
 339              **************************/
 340
 341             r00              = _mm_mul_ps(rsq00,rinv00);
 342             r00              = _mm_andnot_ps(dummy_mask,r00);
 343
 344             /* Compute parameters for interactions between i and j atoms */
 345             qq00             = _mm_mul_ps(iq0,jq0);
 346             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 347                                          vdwparam+vdwioffset0+vdwjidx0B,
 348                                          vdwparam+vdwioffset0+vdwjidx0C,
 349                                          vdwparam+vdwioffset0+vdwjidx0D,
 350                                          &c6_00,&c12_00);
 351
 352             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 353             isaprod          = _mm_mul_ps(isai0,isaj0);
 354             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 355             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 356
 357             /* Calculate generalized born table index - this is a separate table from the normal one,
 358              * but we use the same procedure by multiplying r with scale and truncating to integer.
 359              */
 360             rt               = _mm_mul_ps(r00,gbscale);
 361             gbitab           = _mm_cvttps_epi32(rt);
 362             gbeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(gbitab));
 363             gbitab           = _mm_slli_epi32(gbitab,2);
 364
 365             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 366             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 367             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 368             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 369             _MM_TRANSPOSE4_PS(Y,F,G,H);
 370             Heps             = _mm_mul_ps(gbeps,H);
 371             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 372             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 373             vgb              = _mm_mul_ps(gbqqfactor,VV);
 374
 375             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 376             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 377             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 378             dvdatmp          = _mm_andnot_ps(dummy_mask,dvdatmp);
 379             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 380             /* 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. */
 381             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 382             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 383             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 384             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 385             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 386             velec            = _mm_mul_ps(qq00,rinv00);
 387             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 388
 389             /* LENNARD-JONES DISPERSION/REPULSION */
 390
 391             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 392             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 393             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 394             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 395             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 396
 397             /* Update potential sum for this i atom from the interaction with this j atom. */
 398             velec            = _mm_andnot_ps(dummy_mask,velec);
 399             velecsum         = _mm_add_ps(velecsum,velec);
 400             vgb              = _mm_andnot_ps(dummy_mask,vgb);
 401             vgbsum           = _mm_add_ps(vgbsum,vgb);
 402             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 403             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 404
 405             fscal            = _mm_add_ps(felec,fvdw);
 406
 407             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 408
 409             /* Calculate temporary vectorial force */
 410             tx               = _mm_mul_ps(fscal,dx00);
 411             ty               = _mm_mul_ps(fscal,dy00);
 412             tz               = _mm_mul_ps(fscal,dz00);
 413
 414             /* Update vectorial force */
 415             fix0             = _mm_add_ps(fix0,tx);
 416             fiy0             = _mm_add_ps(fiy0,ty);
 417             fiz0             = _mm_add_ps(fiz0,tz);
 418
 419             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 420             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 421             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 422             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 423             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 424
 425             /* Inner loop uses 72 flops */
 426         }
 427
 428         /* End of innermost loop */
 429
 430         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 431                                               f+i_coord_offset,fshift+i_shift_offset);
 432
 433         ggid                        = gid[iidx];
 434         /* Update potential energies */
 435         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 436         gmx_mm_update_1pot_ps(vgbsum,kernel_data->energygrp_polarization+ggid);
 437         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 438         dvdasum = _mm_mul_ps(dvdasum, _mm_mul_ps(isai0,isai0));
 439         gmx_mm_update_1pot_ps(dvdasum,dvda+inr);
 440
 441         /* Increment number of inner iterations */
 442         inneriter                  += j_index_end - j_index_start;
 443
 444         /* Outer loop uses 10 flops */
 445     }
 446
 447     /* Increment number of outer iterations */
 448     outeriter        += nri;
 449
 450     /* Update outer/inner flops */
 451
 452     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*10 + inneriter*72);
 453 }
 454 /*
 455  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwLJ_GeomP1P1_F_sse2_single
 456  * Electrostatics interaction: GeneralizedBorn
 457  * VdW interaction:            LennardJones
 458  * Geometry:                   Particle-Particle
 459  * Calculate force/pot:        Force
 460  */
 461 void
 462 nb_kernel_ElecGB_VdwLJ_GeomP1P1_F_sse2_single
 463                     (t_nblist * gmx_restrict                nlist,
 464                      rvec * gmx_restrict                    xx,
 465                      rvec * gmx_restrict                    ff,
 466                      t_forcerec * gmx_restrict              fr,
 467                      t_mdatoms * gmx_restrict               mdatoms,
 468                      nb_kernel_data_t * gmx_restrict        kernel_data,
 469                      t_nrnb * gmx_restrict                  nrnb)
 470 {
 471     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 472      * just 0 for non-waters.
 473      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 474      * jnr indices corresponding to data put in the four positions in the SIMD register.
 475      */
 476     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 477     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 478     int              jnrA,jnrB,jnrC,jnrD;
 479     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 480     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 481     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 482     real             rcutoff_scalar;
 483     real             *shiftvec,*fshift,*x,*f;
 484     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 485     real             scratch[4*DIM];
 486     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 487     int              vdwioffset0;
 488     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 489     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 490     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 491     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 492     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 493     real             *charge;
 494     __m128i          gbitab;
 495     __m128           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
 496     __m128           minushalf = _mm_set1_ps(-0.5);
 497     real             *invsqrta,*dvda,*gbtab;
 498     int              nvdwtype;
 499     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 500     int              *vdwtype;
 501     real             *vdwparam;
 502     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 503     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 504     __m128i          vfitab;
 505     __m128i          ifour       = _mm_set1_epi32(4);
 506     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 507     real             *vftab;
 508     __m128           dummy_mask,cutoff_mask;
 509     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 510     __m128           one     = _mm_set1_ps(1.0);
 511     __m128           two     = _mm_set1_ps(2.0);
 512     x                = xx[0];
 513     f                = ff[0];
 514
 515     nri              = nlist->nri;
 516     iinr             = nlist->iinr;
 517     jindex           = nlist->jindex;
 518     jjnr             = nlist->jjnr;
 519     shiftidx         = nlist->shift;
 520     gid              = nlist->gid;
 521     shiftvec         = fr->shift_vec[0];
 522     fshift           = fr->fshift[0];
 523     facel            = _mm_set1_ps(fr->epsfac);
 524     charge           = mdatoms->chargeA;
 525     nvdwtype         = fr->ntype;
 526     vdwparam         = fr->nbfp;
 527     vdwtype          = mdatoms->typeA;
 528
 529     invsqrta         = fr->invsqrta;
 530     dvda             = fr->dvda;
 531     gbtabscale       = _mm_set1_ps(fr->gbtab.scale);
 532     gbtab            = fr->gbtab.data;
 533     gbinvepsdiff     = _mm_set1_ps((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 534
 535     /* Avoid stupid compiler warnings */
 536     jnrA = jnrB = jnrC = jnrD = 0;
 537     j_coord_offsetA = 0;
 538     j_coord_offsetB = 0;
 539     j_coord_offsetC = 0;
 540     j_coord_offsetD = 0;
 541
 542     outeriter        = 0;
 543     inneriter        = 0;
 544
 545     for(iidx=0;iidx<4*DIM;iidx++)
 546     {
 547         scratch[iidx] = 0.0;
 548     }
 549
 550     /* Start outer loop over neighborlists */
 551     for(iidx=0; iidx<nri; iidx++)
 552     {
 553         /* Load shift vector for this list */
 554         i_shift_offset   = DIM*shiftidx[iidx];
 555
 556         /* Load limits for loop over neighbors */
 557         j_index_start    = jindex[iidx];
 558         j_index_end      = jindex[iidx+1];
 559
 560         /* Get outer coordinate index */
 561         inr              = iinr[iidx];
 562         i_coord_offset   = DIM*inr;
 563
 564         /* Load i particle coords and add shift vector */
 565         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 566
 567         fix0             = _mm_setzero_ps();
 568         fiy0             = _mm_setzero_ps();
 569         fiz0             = _mm_setzero_ps();
 570
 571         /* Load parameters for i particles */
 572         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 573         isai0            = _mm_load1_ps(invsqrta+inr+0);
 574         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 575
 576         dvdasum          = _mm_setzero_ps();
 577
 578         /* Start inner kernel loop */
 579         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 580         {
 581
 582             /* Get j neighbor index, and coordinate index */
 583             jnrA             = jjnr[jidx];
 584             jnrB             = jjnr[jidx+1];
 585             jnrC             = jjnr[jidx+2];
 586             jnrD             = jjnr[jidx+3];
 587             j_coord_offsetA  = DIM*jnrA;
 588             j_coord_offsetB  = DIM*jnrB;
 589             j_coord_offsetC  = DIM*jnrC;
 590             j_coord_offsetD  = DIM*jnrD;
 591
 592             /* load j atom coordinates */
 593             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 594                                               x+j_coord_offsetC,x+j_coord_offsetD,
 595                                               &jx0,&jy0,&jz0);
 596
 597             /* Calculate displacement vector */
 598             dx00             = _mm_sub_ps(ix0,jx0);
 599             dy00             = _mm_sub_ps(iy0,jy0);
 600             dz00             = _mm_sub_ps(iz0,jz0);
 601
 602             /* Calculate squared distance and things based on it */
 603             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 604
 605             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 606
 607             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 608
 609             /* Load parameters for j particles */
 610             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 611                                                               charge+jnrC+0,charge+jnrD+0);
 612             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 613                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 614             vdwjidx0A        = 2*vdwtype[jnrA+0];
 615             vdwjidx0B        = 2*vdwtype[jnrB+0];
 616             vdwjidx0C        = 2*vdwtype[jnrC+0];
 617             vdwjidx0D        = 2*vdwtype[jnrD+0];
 618
 619             /**************************
 620              * CALCULATE INTERACTIONS *
 621              **************************/
 622
 623             r00              = _mm_mul_ps(rsq00,rinv00);
 624
 625             /* Compute parameters for interactions between i and j atoms */
 626             qq00             = _mm_mul_ps(iq0,jq0);
 627             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 628                                          vdwparam+vdwioffset0+vdwjidx0B,
 629                                          vdwparam+vdwioffset0+vdwjidx0C,
 630                                          vdwparam+vdwioffset0+vdwjidx0D,
 631                                          &c6_00,&c12_00);
 632
 633             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 634             isaprod          = _mm_mul_ps(isai0,isaj0);
 635             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 636             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 637
 638             /* Calculate generalized born table index - this is a separate table from the normal one,
 639              * but we use the same procedure by multiplying r with scale and truncating to integer.
 640              */
 641             rt               = _mm_mul_ps(r00,gbscale);
 642             gbitab           = _mm_cvttps_epi32(rt);
 643             gbeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(gbitab));
 644             gbitab           = _mm_slli_epi32(gbitab,2);
 645
 646             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 647             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 648             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 649             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 650             _MM_TRANSPOSE4_PS(Y,F,G,H);
 651             Heps             = _mm_mul_ps(gbeps,H);
 652             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 653             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 654             vgb              = _mm_mul_ps(gbqqfactor,VV);
 655
 656             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 657             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 658             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 659             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 660             fjptrA           = dvda+jnrA;
 661             fjptrB           = dvda+jnrB;
 662             fjptrC           = dvda+jnrC;
 663             fjptrD           = dvda+jnrD;
 664             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 665             velec            = _mm_mul_ps(qq00,rinv00);
 666             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 667
 668             /* LENNARD-JONES DISPERSION/REPULSION */
 669
 670             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 671             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 672
 673             fscal            = _mm_add_ps(felec,fvdw);
 674
 675             /* Calculate temporary vectorial force */
 676             tx               = _mm_mul_ps(fscal,dx00);
 677             ty               = _mm_mul_ps(fscal,dy00);
 678             tz               = _mm_mul_ps(fscal,dz00);
 679
 680             /* Update vectorial force */
 681             fix0             = _mm_add_ps(fix0,tx);
 682             fiy0             = _mm_add_ps(fiy0,ty);
 683             fiz0             = _mm_add_ps(fiz0,tz);
 684
 685             fjptrA             = f+j_coord_offsetA;
 686             fjptrB             = f+j_coord_offsetB;
 687             fjptrC             = f+j_coord_offsetC;
 688             fjptrD             = f+j_coord_offsetD;
 689             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 690
 691             /* Inner loop uses 64 flops */
 692         }
 693
 694         if(jidx<j_index_end)
 695         {
 696
 697             /* Get j neighbor index, and coordinate index */
 698             jnrlistA         = jjnr[jidx];
 699             jnrlistB         = jjnr[jidx+1];
 700             jnrlistC         = jjnr[jidx+2];
 701             jnrlistD         = jjnr[jidx+3];
 702             /* Sign of each element will be negative for non-real atoms.
 703              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 704              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 705              */
 706             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 707             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 708             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 709             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 710             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 711             j_coord_offsetA  = DIM*jnrA;
 712             j_coord_offsetB  = DIM*jnrB;
 713             j_coord_offsetC  = DIM*jnrC;
 714             j_coord_offsetD  = DIM*jnrD;
 715
 716             /* load j atom coordinates */
 717             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 718                                               x+j_coord_offsetC,x+j_coord_offsetD,
 719                                               &jx0,&jy0,&jz0);
 720
 721             /* Calculate displacement vector */
 722             dx00             = _mm_sub_ps(ix0,jx0);
 723             dy00             = _mm_sub_ps(iy0,jy0);
 724             dz00             = _mm_sub_ps(iz0,jz0);
 725
 726             /* Calculate squared distance and things based on it */
 727             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 728
 729             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 730
 731             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 732
 733             /* Load parameters for j particles */
 734             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 735                                                               charge+jnrC+0,charge+jnrD+0);
 736             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 737                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 738             vdwjidx0A        = 2*vdwtype[jnrA+0];
 739             vdwjidx0B        = 2*vdwtype[jnrB+0];
 740             vdwjidx0C        = 2*vdwtype[jnrC+0];
 741             vdwjidx0D        = 2*vdwtype[jnrD+0];
 742
 743             /**************************
 744              * CALCULATE INTERACTIONS *
 745              **************************/
 746
 747             r00              = _mm_mul_ps(rsq00,rinv00);
 748             r00              = _mm_andnot_ps(dummy_mask,r00);
 749
 750             /* Compute parameters for interactions between i and j atoms */
 751             qq00             = _mm_mul_ps(iq0,jq0);
 752             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 753                                          vdwparam+vdwioffset0+vdwjidx0B,
 754                                          vdwparam+vdwioffset0+vdwjidx0C,
 755                                          vdwparam+vdwioffset0+vdwjidx0D,
 756                                          &c6_00,&c12_00);
 757
 758             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 759             isaprod          = _mm_mul_ps(isai0,isaj0);
 760             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 761             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 762
 763             /* Calculate generalized born table index - this is a separate table from the normal one,
 764              * but we use the same procedure by multiplying r with scale and truncating to integer.
 765              */
 766             rt               = _mm_mul_ps(r00,gbscale);
 767             gbitab           = _mm_cvttps_epi32(rt);
 768             gbeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(gbitab));
 769             gbitab           = _mm_slli_epi32(gbitab,2);
 770
 771             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 772             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 773             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 774             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 775             _MM_TRANSPOSE4_PS(Y,F,G,H);
 776             Heps             = _mm_mul_ps(gbeps,H);
 777             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 778             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 779             vgb              = _mm_mul_ps(gbqqfactor,VV);
 780
 781             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 782             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 783             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 784             dvdatmp          = _mm_andnot_ps(dummy_mask,dvdatmp);
 785             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 786             /* 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. */
 787             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 788             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 789             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 790             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 791             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 792             velec            = _mm_mul_ps(qq00,rinv00);
 793             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 794
 795             /* LENNARD-JONES DISPERSION/REPULSION */
 796
 797             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 798             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 799
 800             fscal            = _mm_add_ps(felec,fvdw);
 801
 802             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 803
 804             /* Calculate temporary vectorial force */
 805             tx               = _mm_mul_ps(fscal,dx00);
 806             ty               = _mm_mul_ps(fscal,dy00);
 807             tz               = _mm_mul_ps(fscal,dz00);
 808
 809             /* Update vectorial force */
 810             fix0             = _mm_add_ps(fix0,tx);
 811             fiy0             = _mm_add_ps(fiy0,ty);
 812             fiz0             = _mm_add_ps(fiz0,tz);
 813
 814             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 815             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 816             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 817             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 818             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 819
 820             /* Inner loop uses 65 flops */
 821         }
 822
 823         /* End of innermost loop */
 824
 825         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 826                                               f+i_coord_offset,fshift+i_shift_offset);
 827
 828         dvdasum = _mm_mul_ps(dvdasum, _mm_mul_ps(isai0,isai0));
 829         gmx_mm_update_1pot_ps(dvdasum,dvda+inr);
 830
 831         /* Increment number of inner iterations */
 832         inneriter                  += j_index_end - j_index_start;
 833
 834         /* Outer loop uses 7 flops */
 835     }
 836
 837     /* Increment number of outer iterations */
 838     outeriter        += nri;
 839
 840     /* Update outer/inner flops */
 841
 842     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*65);
 843 }