src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_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_ElecEwSh_VdwLJSh_GeomW3P1_VF_sse4_1_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_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              vdwioffset1;
  72     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  73     int              vdwioffset2;
  74     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  75     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  76     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  77     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  78     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  79     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  80     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     int              nvdwtype;
  83     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  84     int              *vdwtype;
  85     real             *vdwparam;
  86     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  87     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  88     __m128i          ewitab;
  89     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  90     real             *ewtab;
  91     __m128           dummy_mask,cutoff_mask;
  92     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  93     __m128           one     = _mm_set1_ps(1.0);
  94     __m128           two     = _mm_set1_ps(2.0);
  95     x                = xx[0];
  96     f                = ff[0];
  97
  98     nri              = nlist->nri;
  99     iinr             = nlist->iinr;
 100     jindex           = nlist->jindex;
 101     jjnr             = nlist->jjnr;
 102     shiftidx         = nlist->shift;
 103     gid              = nlist->gid;
 104     shiftvec         = fr->shift_vec[0];
 105     fshift           = fr->fshift[0];
 106     facel            = _mm_set1_ps(fr->epsfac);
 107     charge           = mdatoms->chargeA;
 108     nvdwtype         = fr->ntype;
 109     vdwparam         = fr->nbfp;
 110     vdwtype          = mdatoms->typeA;
 111
 112     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 113     ewtab            = fr->ic->tabq_coul_FDV0;
 114     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 115     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 116
 117     /* Setup water-specific parameters */
 118     inr              = nlist->iinr[0];
 119     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 120     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 121     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 122     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 123
 124     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 125     rcutoff_scalar   = fr->rcoulomb;
 126     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 127     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 128
 129     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 130     rvdw             = _mm_set1_ps(fr->rvdw);
 131
 132     /* Avoid stupid compiler warnings */
 133     jnrA = jnrB = jnrC = jnrD = 0;
 134     j_coord_offsetA = 0;
 135     j_coord_offsetB = 0;
 136     j_coord_offsetC = 0;
 137     j_coord_offsetD = 0;
 138
 139     outeriter        = 0;
 140     inneriter        = 0;
 141
 142     for(iidx=0;iidx<4*DIM;iidx++)
 143     {
 144         scratch[iidx] = 0.0;
 145     }
 146
 147     /* Start outer loop over neighborlists */
 148     for(iidx=0; iidx<nri; iidx++)
 149     {
 150         /* Load shift vector for this list */
 151         i_shift_offset   = DIM*shiftidx[iidx];
 152
 153         /* Load limits for loop over neighbors */
 154         j_index_start    = jindex[iidx];
 155         j_index_end      = jindex[iidx+1];
 156
 157         /* Get outer coordinate index */
 158         inr              = iinr[iidx];
 159         i_coord_offset   = DIM*inr;
 160
 161         /* Load i particle coords and add shift vector */
 162         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 163                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 164
 165         fix0             = _mm_setzero_ps();
 166         fiy0             = _mm_setzero_ps();
 167         fiz0             = _mm_setzero_ps();
 168         fix1             = _mm_setzero_ps();
 169         fiy1             = _mm_setzero_ps();
 170         fiz1             = _mm_setzero_ps();
 171         fix2             = _mm_setzero_ps();
 172         fiy2             = _mm_setzero_ps();
 173         fiz2             = _mm_setzero_ps();
 174
 175         /* Reset potential sums */
 176         velecsum         = _mm_setzero_ps();
 177         vvdwsum          = _mm_setzero_ps();
 178
 179         /* Start inner kernel loop */
 180         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 181         {
 182
 183             /* Get j neighbor index, and coordinate index */
 184             jnrA             = jjnr[jidx];
 185             jnrB             = jjnr[jidx+1];
 186             jnrC             = jjnr[jidx+2];
 187             jnrD             = jjnr[jidx+3];
 188             j_coord_offsetA  = DIM*jnrA;
 189             j_coord_offsetB  = DIM*jnrB;
 190             j_coord_offsetC  = DIM*jnrC;
 191             j_coord_offsetD  = DIM*jnrD;
 192
 193             /* load j atom coordinates */
 194             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 195                                               x+j_coord_offsetC,x+j_coord_offsetD,
 196                                               &jx0,&jy0,&jz0);
 197
 198             /* Calculate displacement vector */
 199             dx00             = _mm_sub_ps(ix0,jx0);
 200             dy00             = _mm_sub_ps(iy0,jy0);
 201             dz00             = _mm_sub_ps(iz0,jz0);
 202             dx10             = _mm_sub_ps(ix1,jx0);
 203             dy10             = _mm_sub_ps(iy1,jy0);
 204             dz10             = _mm_sub_ps(iz1,jz0);
 205             dx20             = _mm_sub_ps(ix2,jx0);
 206             dy20             = _mm_sub_ps(iy2,jy0);
 207             dz20             = _mm_sub_ps(iz2,jz0);
 208
 209             /* Calculate squared distance and things based on it */
 210             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 211             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 212             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 213
 214             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 215             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 216             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 217
 218             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 219             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 220             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 221
 222             /* Load parameters for j particles */
 223             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 224                                                               charge+jnrC+0,charge+jnrD+0);
 225             vdwjidx0A        = 2*vdwtype[jnrA+0];
 226             vdwjidx0B        = 2*vdwtype[jnrB+0];
 227             vdwjidx0C        = 2*vdwtype[jnrC+0];
 228             vdwjidx0D        = 2*vdwtype[jnrD+0];
 229
 230             /**************************
 231              * CALCULATE INTERACTIONS *
 232              **************************/
 233
 234             if (gmx_mm_any_lt(rsq00,rcutoff2))
 235             {
 236
 237             r00              = _mm_mul_ps(rsq00,rinv00);
 238
 239             /* Compute parameters for interactions between i and j atoms */
 240             qq00             = _mm_mul_ps(iq0,jq0);
 241             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 242                                          vdwparam+vdwioffset0+vdwjidx0B,
 243                                          vdwparam+vdwioffset0+vdwjidx0C,
 244                                          vdwparam+vdwioffset0+vdwjidx0D,
 245                                          &c6_00,&c12_00);
 246
 247             /* EWALD ELECTROSTATICS */
 248
 249             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 250             ewrt             = _mm_mul_ps(r00,ewtabscale);
 251             ewitab           = _mm_cvttps_epi32(ewrt);
 252             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 253             ewitab           = _mm_slli_epi32(ewitab,2);
 254             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 255             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 256             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 257             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 258             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 259             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 260             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 261             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 262             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 263
 264             /* LENNARD-JONES DISPERSION/REPULSION */
 265
 266             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 267             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 268             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 269             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 270                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 271             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 272
 273             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 274
 275             /* Update potential sum for this i atom from the interaction with this j atom. */
 276             velec            = _mm_and_ps(velec,cutoff_mask);
 277             velecsum         = _mm_add_ps(velecsum,velec);
 278             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 279             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 280
 281             fscal            = _mm_add_ps(felec,fvdw);
 282
 283             fscal            = _mm_and_ps(fscal,cutoff_mask);
 284
 285             /* Calculate temporary vectorial force */
 286             tx               = _mm_mul_ps(fscal,dx00);
 287             ty               = _mm_mul_ps(fscal,dy00);
 288             tz               = _mm_mul_ps(fscal,dz00);
 289
 290             /* Update vectorial force */
 291             fix0             = _mm_add_ps(fix0,tx);
 292             fiy0             = _mm_add_ps(fiy0,ty);
 293             fiz0             = _mm_add_ps(fiz0,tz);
 294
 295             fjptrA             = f+j_coord_offsetA;
 296             fjptrB             = f+j_coord_offsetB;
 297             fjptrC             = f+j_coord_offsetC;
 298             fjptrD             = f+j_coord_offsetD;
 299             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 300
 301             }
 302
 303             /**************************
 304              * CALCULATE INTERACTIONS *
 305              **************************/
 306
 307             if (gmx_mm_any_lt(rsq10,rcutoff2))
 308             {
 309
 310             r10              = _mm_mul_ps(rsq10,rinv10);
 311
 312             /* Compute parameters for interactions between i and j atoms */
 313             qq10             = _mm_mul_ps(iq1,jq0);
 314
 315             /* EWALD ELECTROSTATICS */
 316
 317             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 318             ewrt             = _mm_mul_ps(r10,ewtabscale);
 319             ewitab           = _mm_cvttps_epi32(ewrt);
 320             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 321             ewitab           = _mm_slli_epi32(ewitab,2);
 322             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 323             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 324             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 325             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 326             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 327             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 328             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 329             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 330             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 331
 332             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 333
 334             /* Update potential sum for this i atom from the interaction with this j atom. */
 335             velec            = _mm_and_ps(velec,cutoff_mask);
 336             velecsum         = _mm_add_ps(velecsum,velec);
 337
 338             fscal            = felec;
 339
 340             fscal            = _mm_and_ps(fscal,cutoff_mask);
 341
 342             /* Calculate temporary vectorial force */
 343             tx               = _mm_mul_ps(fscal,dx10);
 344             ty               = _mm_mul_ps(fscal,dy10);
 345             tz               = _mm_mul_ps(fscal,dz10);
 346
 347             /* Update vectorial force */
 348             fix1             = _mm_add_ps(fix1,tx);
 349             fiy1             = _mm_add_ps(fiy1,ty);
 350             fiz1             = _mm_add_ps(fiz1,tz);
 351
 352             fjptrA             = f+j_coord_offsetA;
 353             fjptrB             = f+j_coord_offsetB;
 354             fjptrC             = f+j_coord_offsetC;
 355             fjptrD             = f+j_coord_offsetD;
 356             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 357
 358             }
 359
 360             /**************************
 361              * CALCULATE INTERACTIONS *
 362              **************************/
 363
 364             if (gmx_mm_any_lt(rsq20,rcutoff2))
 365             {
 366
 367             r20              = _mm_mul_ps(rsq20,rinv20);
 368
 369             /* Compute parameters for interactions between i and j atoms */
 370             qq20             = _mm_mul_ps(iq2,jq0);
 371
 372             /* EWALD ELECTROSTATICS */
 373
 374             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 375             ewrt             = _mm_mul_ps(r20,ewtabscale);
 376             ewitab           = _mm_cvttps_epi32(ewrt);
 377             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 378             ewitab           = _mm_slli_epi32(ewitab,2);
 379             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 380             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 381             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 382             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 383             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 384             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 385             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 386             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 387             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 388
 389             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 390
 391             /* Update potential sum for this i atom from the interaction with this j atom. */
 392             velec            = _mm_and_ps(velec,cutoff_mask);
 393             velecsum         = _mm_add_ps(velecsum,velec);
 394
 395             fscal            = felec;
 396
 397             fscal            = _mm_and_ps(fscal,cutoff_mask);
 398
 399             /* Calculate temporary vectorial force */
 400             tx               = _mm_mul_ps(fscal,dx20);
 401             ty               = _mm_mul_ps(fscal,dy20);
 402             tz               = _mm_mul_ps(fscal,dz20);
 403
 404             /* Update vectorial force */
 405             fix2             = _mm_add_ps(fix2,tx);
 406             fiy2             = _mm_add_ps(fiy2,ty);
 407             fiz2             = _mm_add_ps(fiz2,tz);
 408
 409             fjptrA             = f+j_coord_offsetA;
 410             fjptrB             = f+j_coord_offsetB;
 411             fjptrC             = f+j_coord_offsetC;
 412             fjptrD             = f+j_coord_offsetD;
 413             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 414
 415             }
 416
 417             /* Inner loop uses 156 flops */
 418         }
 419
 420         if(jidx<j_index_end)
 421         {
 422
 423             /* Get j neighbor index, and coordinate index */
 424             jnrlistA         = jjnr[jidx];
 425             jnrlistB         = jjnr[jidx+1];
 426             jnrlistC         = jjnr[jidx+2];
 427             jnrlistD         = jjnr[jidx+3];
 428             /* Sign of each element will be negative for non-real atoms.
 429              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 430              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 431              */
 432             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 433             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 434             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 435             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 436             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 437             j_coord_offsetA  = DIM*jnrA;
 438             j_coord_offsetB  = DIM*jnrB;
 439             j_coord_offsetC  = DIM*jnrC;
 440             j_coord_offsetD  = DIM*jnrD;
 441
 442             /* load j atom coordinates */
 443             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 444                                               x+j_coord_offsetC,x+j_coord_offsetD,
 445                                               &jx0,&jy0,&jz0);
 446
 447             /* Calculate displacement vector */
 448             dx00             = _mm_sub_ps(ix0,jx0);
 449             dy00             = _mm_sub_ps(iy0,jy0);
 450             dz00             = _mm_sub_ps(iz0,jz0);
 451             dx10             = _mm_sub_ps(ix1,jx0);
 452             dy10             = _mm_sub_ps(iy1,jy0);
 453             dz10             = _mm_sub_ps(iz1,jz0);
 454             dx20             = _mm_sub_ps(ix2,jx0);
 455             dy20             = _mm_sub_ps(iy2,jy0);
 456             dz20             = _mm_sub_ps(iz2,jz0);
 457
 458             /* Calculate squared distance and things based on it */
 459             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 460             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 461             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 462
 463             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 464             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 465             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 466
 467             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 468             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 469             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 470
 471             /* Load parameters for j particles */
 472             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 473                                                               charge+jnrC+0,charge+jnrD+0);
 474             vdwjidx0A        = 2*vdwtype[jnrA+0];
 475             vdwjidx0B        = 2*vdwtype[jnrB+0];
 476             vdwjidx0C        = 2*vdwtype[jnrC+0];
 477             vdwjidx0D        = 2*vdwtype[jnrD+0];
 478
 479             /**************************
 480              * CALCULATE INTERACTIONS *
 481              **************************/
 482
 483             if (gmx_mm_any_lt(rsq00,rcutoff2))
 484             {
 485
 486             r00              = _mm_mul_ps(rsq00,rinv00);
 487             r00              = _mm_andnot_ps(dummy_mask,r00);
 488
 489             /* Compute parameters for interactions between i and j atoms */
 490             qq00             = _mm_mul_ps(iq0,jq0);
 491             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 492                                          vdwparam+vdwioffset0+vdwjidx0B,
 493                                          vdwparam+vdwioffset0+vdwjidx0C,
 494                                          vdwparam+vdwioffset0+vdwjidx0D,
 495                                          &c6_00,&c12_00);
 496
 497             /* EWALD ELECTROSTATICS */
 498
 499             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 500             ewrt             = _mm_mul_ps(r00,ewtabscale);
 501             ewitab           = _mm_cvttps_epi32(ewrt);
 502             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 503             ewitab           = _mm_slli_epi32(ewitab,2);
 504             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 505             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 506             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 507             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 508             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 509             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 510             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 511             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 512             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 513
 514             /* LENNARD-JONES DISPERSION/REPULSION */
 515
 516             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 517             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 518             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 519             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 520                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 521             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 522
 523             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 524
 525             /* Update potential sum for this i atom from the interaction with this j atom. */
 526             velec            = _mm_and_ps(velec,cutoff_mask);
 527             velec            = _mm_andnot_ps(dummy_mask,velec);
 528             velecsum         = _mm_add_ps(velecsum,velec);
 529             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 530             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 531             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 532
 533             fscal            = _mm_add_ps(felec,fvdw);
 534
 535             fscal            = _mm_and_ps(fscal,cutoff_mask);
 536
 537             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 538
 539             /* Calculate temporary vectorial force */
 540             tx               = _mm_mul_ps(fscal,dx00);
 541             ty               = _mm_mul_ps(fscal,dy00);
 542             tz               = _mm_mul_ps(fscal,dz00);
 543
 544             /* Update vectorial force */
 545             fix0             = _mm_add_ps(fix0,tx);
 546             fiy0             = _mm_add_ps(fiy0,ty);
 547             fiz0             = _mm_add_ps(fiz0,tz);
 548
 549             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 550             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 551             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 552             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 553             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 554
 555             }
 556
 557             /**************************
 558              * CALCULATE INTERACTIONS *
 559              **************************/
 560
 561             if (gmx_mm_any_lt(rsq10,rcutoff2))
 562             {
 563
 564             r10              = _mm_mul_ps(rsq10,rinv10);
 565             r10              = _mm_andnot_ps(dummy_mask,r10);
 566
 567             /* Compute parameters for interactions between i and j atoms */
 568             qq10             = _mm_mul_ps(iq1,jq0);
 569
 570             /* EWALD ELECTROSTATICS */
 571
 572             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 573             ewrt             = _mm_mul_ps(r10,ewtabscale);
 574             ewitab           = _mm_cvttps_epi32(ewrt);
 575             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 576             ewitab           = _mm_slli_epi32(ewitab,2);
 577             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 578             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 579             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 580             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 581             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 582             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 583             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 584             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 585             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 586
 587             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 588
 589             /* Update potential sum for this i atom from the interaction with this j atom. */
 590             velec            = _mm_and_ps(velec,cutoff_mask);
 591             velec            = _mm_andnot_ps(dummy_mask,velec);
 592             velecsum         = _mm_add_ps(velecsum,velec);
 593
 594             fscal            = felec;
 595
 596             fscal            = _mm_and_ps(fscal,cutoff_mask);
 597
 598             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 599
 600             /* Calculate temporary vectorial force */
 601             tx               = _mm_mul_ps(fscal,dx10);
 602             ty               = _mm_mul_ps(fscal,dy10);
 603             tz               = _mm_mul_ps(fscal,dz10);
 604
 605             /* Update vectorial force */
 606             fix1             = _mm_add_ps(fix1,tx);
 607             fiy1             = _mm_add_ps(fiy1,ty);
 608             fiz1             = _mm_add_ps(fiz1,tz);
 609
 610             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 611             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 612             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 613             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 614             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 615
 616             }
 617
 618             /**************************
 619              * CALCULATE INTERACTIONS *
 620              **************************/
 621
 622             if (gmx_mm_any_lt(rsq20,rcutoff2))
 623             {
 624
 625             r20              = _mm_mul_ps(rsq20,rinv20);
 626             r20              = _mm_andnot_ps(dummy_mask,r20);
 627
 628             /* Compute parameters for interactions between i and j atoms */
 629             qq20             = _mm_mul_ps(iq2,jq0);
 630
 631             /* EWALD ELECTROSTATICS */
 632
 633             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 634             ewrt             = _mm_mul_ps(r20,ewtabscale);
 635             ewitab           = _mm_cvttps_epi32(ewrt);
 636             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 637             ewitab           = _mm_slli_epi32(ewitab,2);
 638             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 639             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 640             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 641             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 642             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 643             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 644             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 645             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 646             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 647
 648             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 649
 650             /* Update potential sum for this i atom from the interaction with this j atom. */
 651             velec            = _mm_and_ps(velec,cutoff_mask);
 652             velec            = _mm_andnot_ps(dummy_mask,velec);
 653             velecsum         = _mm_add_ps(velecsum,velec);
 654
 655             fscal            = felec;
 656
 657             fscal            = _mm_and_ps(fscal,cutoff_mask);
 658
 659             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 660
 661             /* Calculate temporary vectorial force */
 662             tx               = _mm_mul_ps(fscal,dx20);
 663             ty               = _mm_mul_ps(fscal,dy20);
 664             tz               = _mm_mul_ps(fscal,dz20);
 665
 666             /* Update vectorial force */
 667             fix2             = _mm_add_ps(fix2,tx);
 668             fiy2             = _mm_add_ps(fiy2,ty);
 669             fiz2             = _mm_add_ps(fiz2,tz);
 670
 671             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 672             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 673             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 674             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 675             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 676
 677             }
 678
 679             /* Inner loop uses 159 flops */
 680         }
 681
 682         /* End of innermost loop */
 683
 684         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 685                                               f+i_coord_offset,fshift+i_shift_offset);
 686
 687         ggid                        = gid[iidx];
 688         /* Update potential energies */
 689         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 690         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 691
 692         /* Increment number of inner iterations */
 693         inneriter                  += j_index_end - j_index_start;
 694
 695         /* Outer loop uses 20 flops */
 696     }
 697
 698     /* Increment number of outer iterations */
 699     outeriter        += nri;
 700
 701     /* Update outer/inner flops */
 702
 703     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*159);
 704 }
 705 /*
 706  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_sse4_1_single
 707  * Electrostatics interaction: Ewald
 708  * VdW interaction:            LennardJones
 709  * Geometry:                   Water3-Particle
 710  * Calculate force/pot:        Force
 711  */
 712 void
 713 nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_sse4_1_single
 714                     (t_nblist * gmx_restrict                nlist,
 715                      rvec * gmx_restrict                    xx,
 716                      rvec * gmx_restrict                    ff,
 717                      t_forcerec * gmx_restrict              fr,
 718                      t_mdatoms * gmx_restrict               mdatoms,
 719                      nb_kernel_data_t * gmx_restrict        kernel_data,
 720                      t_nrnb * gmx_restrict                  nrnb)
 721 {
 722     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 723      * just 0 for non-waters.
 724      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 725      * jnr indices corresponding to data put in the four positions in the SIMD register.
 726      */
 727     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 728     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 729     int              jnrA,jnrB,jnrC,jnrD;
 730     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 731     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 732     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 733     real             rcutoff_scalar;
 734     real             *shiftvec,*fshift,*x,*f;
 735     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 736     real             scratch[4*DIM];
 737     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 738     int              vdwioffset0;
 739     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 740     int              vdwioffset1;
 741     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 742     int              vdwioffset2;
 743     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 744     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 745     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 746     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 747     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 748     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 749     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 750     real             *charge;
 751     int              nvdwtype;
 752     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 753     int              *vdwtype;
 754     real             *vdwparam;
 755     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 756     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 757     __m128i          ewitab;
 758     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 759     real             *ewtab;
 760     __m128           dummy_mask,cutoff_mask;
 761     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 762     __m128           one     = _mm_set1_ps(1.0);
 763     __m128           two     = _mm_set1_ps(2.0);
 764     x                = xx[0];
 765     f                = ff[0];
 766
 767     nri              = nlist->nri;
 768     iinr             = nlist->iinr;
 769     jindex           = nlist->jindex;
 770     jjnr             = nlist->jjnr;
 771     shiftidx         = nlist->shift;
 772     gid              = nlist->gid;
 773     shiftvec         = fr->shift_vec[0];
 774     fshift           = fr->fshift[0];
 775     facel            = _mm_set1_ps(fr->epsfac);
 776     charge           = mdatoms->chargeA;
 777     nvdwtype         = fr->ntype;
 778     vdwparam         = fr->nbfp;
 779     vdwtype          = mdatoms->typeA;
 780
 781     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 782     ewtab            = fr->ic->tabq_coul_F;
 783     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 784     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 785
 786     /* Setup water-specific parameters */
 787     inr              = nlist->iinr[0];
 788     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 789     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 790     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 791     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 792
 793     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 794     rcutoff_scalar   = fr->rcoulomb;
 795     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 796     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 797
 798     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 799     rvdw             = _mm_set1_ps(fr->rvdw);
 800
 801     /* Avoid stupid compiler warnings */
 802     jnrA = jnrB = jnrC = jnrD = 0;
 803     j_coord_offsetA = 0;
 804     j_coord_offsetB = 0;
 805     j_coord_offsetC = 0;
 806     j_coord_offsetD = 0;
 807
 808     outeriter        = 0;
 809     inneriter        = 0;
 810
 811     for(iidx=0;iidx<4*DIM;iidx++)
 812     {
 813         scratch[iidx] = 0.0;
 814     }
 815
 816     /* Start outer loop over neighborlists */
 817     for(iidx=0; iidx<nri; iidx++)
 818     {
 819         /* Load shift vector for this list */
 820         i_shift_offset   = DIM*shiftidx[iidx];
 821
 822         /* Load limits for loop over neighbors */
 823         j_index_start    = jindex[iidx];
 824         j_index_end      = jindex[iidx+1];
 825
 826         /* Get outer coordinate index */
 827         inr              = iinr[iidx];
 828         i_coord_offset   = DIM*inr;
 829
 830         /* Load i particle coords and add shift vector */
 831         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 832                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 833
 834         fix0             = _mm_setzero_ps();
 835         fiy0             = _mm_setzero_ps();
 836         fiz0             = _mm_setzero_ps();
 837         fix1             = _mm_setzero_ps();
 838         fiy1             = _mm_setzero_ps();
 839         fiz1             = _mm_setzero_ps();
 840         fix2             = _mm_setzero_ps();
 841         fiy2             = _mm_setzero_ps();
 842         fiz2             = _mm_setzero_ps();
 843
 844         /* Start inner kernel loop */
 845         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 846         {
 847
 848             /* Get j neighbor index, and coordinate index */
 849             jnrA             = jjnr[jidx];
 850             jnrB             = jjnr[jidx+1];
 851             jnrC             = jjnr[jidx+2];
 852             jnrD             = jjnr[jidx+3];
 853             j_coord_offsetA  = DIM*jnrA;
 854             j_coord_offsetB  = DIM*jnrB;
 855             j_coord_offsetC  = DIM*jnrC;
 856             j_coord_offsetD  = DIM*jnrD;
 857
 858             /* load j atom coordinates */
 859             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 860                                               x+j_coord_offsetC,x+j_coord_offsetD,
 861                                               &jx0,&jy0,&jz0);
 862
 863             /* Calculate displacement vector */
 864             dx00             = _mm_sub_ps(ix0,jx0);
 865             dy00             = _mm_sub_ps(iy0,jy0);
 866             dz00             = _mm_sub_ps(iz0,jz0);
 867             dx10             = _mm_sub_ps(ix1,jx0);
 868             dy10             = _mm_sub_ps(iy1,jy0);
 869             dz10             = _mm_sub_ps(iz1,jz0);
 870             dx20             = _mm_sub_ps(ix2,jx0);
 871             dy20             = _mm_sub_ps(iy2,jy0);
 872             dz20             = _mm_sub_ps(iz2,jz0);
 873
 874             /* Calculate squared distance and things based on it */
 875             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 876             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 877             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 878
 879             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 880             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 881             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 882
 883             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 884             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 885             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 886
 887             /* Load parameters for j particles */
 888             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 889                                                               charge+jnrC+0,charge+jnrD+0);
 890             vdwjidx0A        = 2*vdwtype[jnrA+0];
 891             vdwjidx0B        = 2*vdwtype[jnrB+0];
 892             vdwjidx0C        = 2*vdwtype[jnrC+0];
 893             vdwjidx0D        = 2*vdwtype[jnrD+0];
 894
 895             /**************************
 896              * CALCULATE INTERACTIONS *
 897              **************************/
 898
 899             if (gmx_mm_any_lt(rsq00,rcutoff2))
 900             {
 901
 902             r00              = _mm_mul_ps(rsq00,rinv00);
 903
 904             /* Compute parameters for interactions between i and j atoms */
 905             qq00             = _mm_mul_ps(iq0,jq0);
 906             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 907                                          vdwparam+vdwioffset0+vdwjidx0B,
 908                                          vdwparam+vdwioffset0+vdwjidx0C,
 909                                          vdwparam+vdwioffset0+vdwjidx0D,
 910                                          &c6_00,&c12_00);
 911
 912             /* EWALD ELECTROSTATICS */
 913
 914             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 915             ewrt             = _mm_mul_ps(r00,ewtabscale);
 916             ewitab           = _mm_cvttps_epi32(ewrt);
 917             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 918             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 919                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 920                                          &ewtabF,&ewtabFn);
 921             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 922             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 923
 924             /* LENNARD-JONES DISPERSION/REPULSION */
 925
 926             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 927             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 928
 929             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 930
 931             fscal            = _mm_add_ps(felec,fvdw);
 932
 933             fscal            = _mm_and_ps(fscal,cutoff_mask);
 934
 935             /* Calculate temporary vectorial force */
 936             tx               = _mm_mul_ps(fscal,dx00);
 937             ty               = _mm_mul_ps(fscal,dy00);
 938             tz               = _mm_mul_ps(fscal,dz00);
 939
 940             /* Update vectorial force */
 941             fix0             = _mm_add_ps(fix0,tx);
 942             fiy0             = _mm_add_ps(fiy0,ty);
 943             fiz0             = _mm_add_ps(fiz0,tz);
 944
 945             fjptrA             = f+j_coord_offsetA;
 946             fjptrB             = f+j_coord_offsetB;
 947             fjptrC             = f+j_coord_offsetC;
 948             fjptrD             = f+j_coord_offsetD;
 949             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 950
 951             }
 952
 953             /**************************
 954              * CALCULATE INTERACTIONS *
 955              **************************/
 956
 957             if (gmx_mm_any_lt(rsq10,rcutoff2))
 958             {
 959
 960             r10              = _mm_mul_ps(rsq10,rinv10);
 961
 962             /* Compute parameters for interactions between i and j atoms */
 963             qq10             = _mm_mul_ps(iq1,jq0);
 964
 965             /* EWALD ELECTROSTATICS */
 966
 967             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 968             ewrt             = _mm_mul_ps(r10,ewtabscale);
 969             ewitab           = _mm_cvttps_epi32(ewrt);
 970             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 971             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 972                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 973                                          &ewtabF,&ewtabFn);
 974             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 975             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 976
 977             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 978
 979             fscal            = felec;
 980
 981             fscal            = _mm_and_ps(fscal,cutoff_mask);
 982
 983             /* Calculate temporary vectorial force */
 984             tx               = _mm_mul_ps(fscal,dx10);
 985             ty               = _mm_mul_ps(fscal,dy10);
 986             tz               = _mm_mul_ps(fscal,dz10);
 987
 988             /* Update vectorial force */
 989             fix1             = _mm_add_ps(fix1,tx);
 990             fiy1             = _mm_add_ps(fiy1,ty);
 991             fiz1             = _mm_add_ps(fiz1,tz);
 992
 993             fjptrA             = f+j_coord_offsetA;
 994             fjptrB             = f+j_coord_offsetB;
 995             fjptrC             = f+j_coord_offsetC;
 996             fjptrD             = f+j_coord_offsetD;
 997             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 998
 999             }
1000
1001             /**************************
1002              * CALCULATE INTERACTIONS *
1003              **************************/
1004
1005             if (gmx_mm_any_lt(rsq20,rcutoff2))
1006             {
1007
1008             r20              = _mm_mul_ps(rsq20,rinv20);
1009
1010             /* Compute parameters for interactions between i and j atoms */
1011             qq20             = _mm_mul_ps(iq2,jq0);
1012
1013             /* EWALD ELECTROSTATICS */
1014
1015             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1016             ewrt             = _mm_mul_ps(r20,ewtabscale);
1017             ewitab           = _mm_cvttps_epi32(ewrt);
1018             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1019             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1020                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1021                                          &ewtabF,&ewtabFn);
1022             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1023             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1024
1025             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1026
1027             fscal            = felec;
1028
1029             fscal            = _mm_and_ps(fscal,cutoff_mask);
1030
1031             /* Calculate temporary vectorial force */
1032             tx               = _mm_mul_ps(fscal,dx20);
1033             ty               = _mm_mul_ps(fscal,dy20);
1034             tz               = _mm_mul_ps(fscal,dz20);
1035
1036             /* Update vectorial force */
1037             fix2             = _mm_add_ps(fix2,tx);
1038             fiy2             = _mm_add_ps(fiy2,ty);
1039             fiz2             = _mm_add_ps(fiz2,tz);
1040
1041             fjptrA             = f+j_coord_offsetA;
1042             fjptrB             = f+j_coord_offsetB;
1043             fjptrC             = f+j_coord_offsetC;
1044             fjptrD             = f+j_coord_offsetD;
1045             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1046
1047             }
1048
1049             /* Inner loop uses 124 flops */
1050         }
1051
1052         if(jidx<j_index_end)
1053         {
1054
1055             /* Get j neighbor index, and coordinate index */
1056             jnrlistA         = jjnr[jidx];
1057             jnrlistB         = jjnr[jidx+1];
1058             jnrlistC         = jjnr[jidx+2];
1059             jnrlistD         = jjnr[jidx+3];
1060             /* Sign of each element will be negative for non-real atoms.
1061              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1062              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1063              */
1064             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1065             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1066             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1067             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1068             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1069             j_coord_offsetA  = DIM*jnrA;
1070             j_coord_offsetB  = DIM*jnrB;
1071             j_coord_offsetC  = DIM*jnrC;
1072             j_coord_offsetD  = DIM*jnrD;
1073
1074             /* load j atom coordinates */
1075             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1076                                               x+j_coord_offsetC,x+j_coord_offsetD,
1077                                               &jx0,&jy0,&jz0);
1078
1079             /* Calculate displacement vector */
1080             dx00             = _mm_sub_ps(ix0,jx0);
1081             dy00             = _mm_sub_ps(iy0,jy0);
1082             dz00             = _mm_sub_ps(iz0,jz0);
1083             dx10             = _mm_sub_ps(ix1,jx0);
1084             dy10             = _mm_sub_ps(iy1,jy0);
1085             dz10             = _mm_sub_ps(iz1,jz0);
1086             dx20             = _mm_sub_ps(ix2,jx0);
1087             dy20             = _mm_sub_ps(iy2,jy0);
1088             dz20             = _mm_sub_ps(iz2,jz0);
1089
1090             /* Calculate squared distance and things based on it */
1091             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1092             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1093             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1094
1095             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1096             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1097             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1098
1099             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1100             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1101             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1102
1103             /* Load parameters for j particles */
1104             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1105                                                               charge+jnrC+0,charge+jnrD+0);
1106             vdwjidx0A        = 2*vdwtype[jnrA+0];
1107             vdwjidx0B        = 2*vdwtype[jnrB+0];
1108             vdwjidx0C        = 2*vdwtype[jnrC+0];
1109             vdwjidx0D        = 2*vdwtype[jnrD+0];
1110
1111             /**************************
1112              * CALCULATE INTERACTIONS *
1113              **************************/
1114
1115             if (gmx_mm_any_lt(rsq00,rcutoff2))
1116             {
1117
1118             r00              = _mm_mul_ps(rsq00,rinv00);
1119             r00              = _mm_andnot_ps(dummy_mask,r00);
1120
1121             /* Compute parameters for interactions between i and j atoms */
1122             qq00             = _mm_mul_ps(iq0,jq0);
1123             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1124                                          vdwparam+vdwioffset0+vdwjidx0B,
1125                                          vdwparam+vdwioffset0+vdwjidx0C,
1126                                          vdwparam+vdwioffset0+vdwjidx0D,
1127                                          &c6_00,&c12_00);
1128
1129             /* EWALD ELECTROSTATICS */
1130
1131             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1132             ewrt             = _mm_mul_ps(r00,ewtabscale);
1133             ewitab           = _mm_cvttps_epi32(ewrt);
1134             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1135             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1136                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1137                                          &ewtabF,&ewtabFn);
1138             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1139             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1140
1141             /* LENNARD-JONES DISPERSION/REPULSION */
1142
1143             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1144             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1145
1146             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1147
1148             fscal            = _mm_add_ps(felec,fvdw);
1149
1150             fscal            = _mm_and_ps(fscal,cutoff_mask);
1151
1152             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1153
1154             /* Calculate temporary vectorial force */
1155             tx               = _mm_mul_ps(fscal,dx00);
1156             ty               = _mm_mul_ps(fscal,dy00);
1157             tz               = _mm_mul_ps(fscal,dz00);
1158
1159             /* Update vectorial force */
1160             fix0             = _mm_add_ps(fix0,tx);
1161             fiy0             = _mm_add_ps(fiy0,ty);
1162             fiz0             = _mm_add_ps(fiz0,tz);
1163
1164             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1165             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1166             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1167             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1168             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1169
1170             }
1171
1172             /**************************
1173              * CALCULATE INTERACTIONS *
1174              **************************/
1175
1176             if (gmx_mm_any_lt(rsq10,rcutoff2))
1177             {
1178
1179             r10              = _mm_mul_ps(rsq10,rinv10);
1180             r10              = _mm_andnot_ps(dummy_mask,r10);
1181
1182             /* Compute parameters for interactions between i and j atoms */
1183             qq10             = _mm_mul_ps(iq1,jq0);
1184
1185             /* EWALD ELECTROSTATICS */
1186
1187             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1188             ewrt             = _mm_mul_ps(r10,ewtabscale);
1189             ewitab           = _mm_cvttps_epi32(ewrt);
1190             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1191             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1192                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1193                                          &ewtabF,&ewtabFn);
1194             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1195             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1196
1197             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1198
1199             fscal            = felec;
1200
1201             fscal            = _mm_and_ps(fscal,cutoff_mask);
1202
1203             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1204
1205             /* Calculate temporary vectorial force */
1206             tx               = _mm_mul_ps(fscal,dx10);
1207             ty               = _mm_mul_ps(fscal,dy10);
1208             tz               = _mm_mul_ps(fscal,dz10);
1209
1210             /* Update vectorial force */
1211             fix1             = _mm_add_ps(fix1,tx);
1212             fiy1             = _mm_add_ps(fiy1,ty);
1213             fiz1             = _mm_add_ps(fiz1,tz);
1214
1215             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1216             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1217             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1218             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1219             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1220
1221             }
1222
1223             /**************************
1224              * CALCULATE INTERACTIONS *
1225              **************************/
1226
1227             if (gmx_mm_any_lt(rsq20,rcutoff2))
1228             {
1229
1230             r20              = _mm_mul_ps(rsq20,rinv20);
1231             r20              = _mm_andnot_ps(dummy_mask,r20);
1232
1233             /* Compute parameters for interactions between i and j atoms */
1234             qq20             = _mm_mul_ps(iq2,jq0);
1235
1236             /* EWALD ELECTROSTATICS */
1237
1238             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1239             ewrt             = _mm_mul_ps(r20,ewtabscale);
1240             ewitab           = _mm_cvttps_epi32(ewrt);
1241             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1242             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1243                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1244                                          &ewtabF,&ewtabFn);
1245             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1246             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1247
1248             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1249
1250             fscal            = felec;
1251
1252             fscal            = _mm_and_ps(fscal,cutoff_mask);
1253
1254             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1255
1256             /* Calculate temporary vectorial force */
1257             tx               = _mm_mul_ps(fscal,dx20);
1258             ty               = _mm_mul_ps(fscal,dy20);
1259             tz               = _mm_mul_ps(fscal,dz20);
1260
1261             /* Update vectorial force */
1262             fix2             = _mm_add_ps(fix2,tx);
1263             fiy2             = _mm_add_ps(fiy2,ty);
1264             fiz2             = _mm_add_ps(fiz2,tz);
1265
1266             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1267             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1268             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1269             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1270             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1271
1272             }
1273
1274             /* Inner loop uses 127 flops */
1275         }
1276
1277         /* End of innermost loop */
1278
1279         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1280                                               f+i_coord_offset,fshift+i_shift_offset);
1281
1282         /* Increment number of inner iterations */
1283         inneriter                  += j_index_end - j_index_start;
1284
1285         /* Outer loop uses 18 flops */
1286     }
1287
1288     /* Increment number of outer iterations */
1289     outeriter        += nri;
1290
1291     /* Update outer/inner flops */
1292
1293     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*127);
1294 }