src/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_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_ElecRFCut_VdwCSTab_GeomW4P1_VF_sse4_1_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_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              vdwioffset3;
  76     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  77     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  78     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  79     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  80     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  81     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  82     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  83     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  84     real             *charge;
  85     int              nvdwtype;
  86     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  87     int              *vdwtype;
  88     real             *vdwparam;
  89     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  90     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  91     __m128i          vfitab;
  92     __m128i          ifour       = _mm_set1_epi32(4);
  93     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  94     real             *vftab;
  95     __m128           dummy_mask,cutoff_mask;
  96     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  97     __m128           one     = _mm_set1_ps(1.0);
  98     __m128           two     = _mm_set1_ps(2.0);
  99     x                = xx[0];
 100     f                = ff[0];
 101
 102     nri              = nlist->nri;
 103     iinr             = nlist->iinr;
 104     jindex           = nlist->jindex;
 105     jjnr             = nlist->jjnr;
 106     shiftidx         = nlist->shift;
 107     gid              = nlist->gid;
 108     shiftvec         = fr->shift_vec[0];
 109     fshift           = fr->fshift[0];
 110     facel            = _mm_set1_ps(fr->epsfac);
 111     charge           = mdatoms->chargeA;
 112     krf              = _mm_set1_ps(fr->ic->k_rf);
 113     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 114     crf              = _mm_set1_ps(fr->ic->c_rf);
 115     nvdwtype         = fr->ntype;
 116     vdwparam         = fr->nbfp;
 117     vdwtype          = mdatoms->typeA;
 118
 119     vftab            = kernel_data->table_vdw->data;
 120     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 121
 122     /* Setup water-specific parameters */
 123     inr              = nlist->iinr[0];
 124     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 125     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 126     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 127     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 128
 129     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 130     rcutoff_scalar   = fr->rcoulomb;
 131     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 132     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 133
 134     /* Avoid stupid compiler warnings */
 135     jnrA = jnrB = jnrC = jnrD = 0;
 136     j_coord_offsetA = 0;
 137     j_coord_offsetB = 0;
 138     j_coord_offsetC = 0;
 139     j_coord_offsetD = 0;
 140
 141     outeriter        = 0;
 142     inneriter        = 0;
 143
 144     for(iidx=0;iidx<4*DIM;iidx++)
 145     {
 146         scratch[iidx] = 0.0;
 147     }
 148
 149     /* Start outer loop over neighborlists */
 150     for(iidx=0; iidx<nri; iidx++)
 151     {
 152         /* Load shift vector for this list */
 153         i_shift_offset   = DIM*shiftidx[iidx];
 154
 155         /* Load limits for loop over neighbors */
 156         j_index_start    = jindex[iidx];
 157         j_index_end      = jindex[iidx+1];
 158
 159         /* Get outer coordinate index */
 160         inr              = iinr[iidx];
 161         i_coord_offset   = DIM*inr;
 162
 163         /* Load i particle coords and add shift vector */
 164         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 165                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 166
 167         fix0             = _mm_setzero_ps();
 168         fiy0             = _mm_setzero_ps();
 169         fiz0             = _mm_setzero_ps();
 170         fix1             = _mm_setzero_ps();
 171         fiy1             = _mm_setzero_ps();
 172         fiz1             = _mm_setzero_ps();
 173         fix2             = _mm_setzero_ps();
 174         fiy2             = _mm_setzero_ps();
 175         fiz2             = _mm_setzero_ps();
 176         fix3             = _mm_setzero_ps();
 177         fiy3             = _mm_setzero_ps();
 178         fiz3             = _mm_setzero_ps();
 179
 180         /* Reset potential sums */
 181         velecsum         = _mm_setzero_ps();
 182         vvdwsum          = _mm_setzero_ps();
 183
 184         /* Start inner kernel loop */
 185         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 186         {
 187
 188             /* Get j neighbor index, and coordinate index */
 189             jnrA             = jjnr[jidx];
 190             jnrB             = jjnr[jidx+1];
 191             jnrC             = jjnr[jidx+2];
 192             jnrD             = jjnr[jidx+3];
 193             j_coord_offsetA  = DIM*jnrA;
 194             j_coord_offsetB  = DIM*jnrB;
 195             j_coord_offsetC  = DIM*jnrC;
 196             j_coord_offsetD  = DIM*jnrD;
 197
 198             /* load j atom coordinates */
 199             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 200                                               x+j_coord_offsetC,x+j_coord_offsetD,
 201                                               &jx0,&jy0,&jz0);
 202
 203             /* Calculate displacement vector */
 204             dx00             = _mm_sub_ps(ix0,jx0);
 205             dy00             = _mm_sub_ps(iy0,jy0);
 206             dz00             = _mm_sub_ps(iz0,jz0);
 207             dx10             = _mm_sub_ps(ix1,jx0);
 208             dy10             = _mm_sub_ps(iy1,jy0);
 209             dz10             = _mm_sub_ps(iz1,jz0);
 210             dx20             = _mm_sub_ps(ix2,jx0);
 211             dy20             = _mm_sub_ps(iy2,jy0);
 212             dz20             = _mm_sub_ps(iz2,jz0);
 213             dx30             = _mm_sub_ps(ix3,jx0);
 214             dy30             = _mm_sub_ps(iy3,jy0);
 215             dz30             = _mm_sub_ps(iz3,jz0);
 216
 217             /* Calculate squared distance and things based on it */
 218             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 219             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 220             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 221             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 222
 223             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 224             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 225             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 226             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 227
 228             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 229             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 230             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 231
 232             /* Load parameters for j particles */
 233             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 234                                                               charge+jnrC+0,charge+jnrD+0);
 235             vdwjidx0A        = 2*vdwtype[jnrA+0];
 236             vdwjidx0B        = 2*vdwtype[jnrB+0];
 237             vdwjidx0C        = 2*vdwtype[jnrC+0];
 238             vdwjidx0D        = 2*vdwtype[jnrD+0];
 239
 240             fjx0             = _mm_setzero_ps();
 241             fjy0             = _mm_setzero_ps();
 242             fjz0             = _mm_setzero_ps();
 243
 244             /**************************
 245              * CALCULATE INTERACTIONS *
 246              **************************/
 247
 248             r00              = _mm_mul_ps(rsq00,rinv00);
 249
 250             /* Compute parameters for interactions between i and j atoms */
 251             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 252                                          vdwparam+vdwioffset0+vdwjidx0B,
 253                                          vdwparam+vdwioffset0+vdwjidx0C,
 254                                          vdwparam+vdwioffset0+vdwjidx0D,
 255                                          &c6_00,&c12_00);
 256
 257             /* Calculate table index by multiplying r with table scale and truncate to integer */
 258             rt               = _mm_mul_ps(r00,vftabscale);
 259             vfitab           = _mm_cvttps_epi32(rt);
 260             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 261             vfitab           = _mm_slli_epi32(vfitab,3);
 262
 263             /* CUBIC SPLINE TABLE DISPERSION */
 264             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 265             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 266             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 267             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 268             _MM_TRANSPOSE4_PS(Y,F,G,H);
 269             Heps             = _mm_mul_ps(vfeps,H);
 270             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 271             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 272             vvdw6            = _mm_mul_ps(c6_00,VV);
 273             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 274             fvdw6            = _mm_mul_ps(c6_00,FF);
 275
 276             /* CUBIC SPLINE TABLE REPULSION */
 277             vfitab           = _mm_add_epi32(vfitab,ifour);
 278             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 279             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 280             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 281             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 282             _MM_TRANSPOSE4_PS(Y,F,G,H);
 283             Heps             = _mm_mul_ps(vfeps,H);
 284             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 285             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 286             vvdw12           = _mm_mul_ps(c12_00,VV);
 287             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 288             fvdw12           = _mm_mul_ps(c12_00,FF);
 289             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 290             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 291
 292             /* Update potential sum for this i atom from the interaction with this j atom. */
 293             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 294
 295             fscal            = fvdw;
 296
 297             /* Calculate temporary vectorial force */
 298             tx               = _mm_mul_ps(fscal,dx00);
 299             ty               = _mm_mul_ps(fscal,dy00);
 300             tz               = _mm_mul_ps(fscal,dz00);
 301
 302             /* Update vectorial force */
 303             fix0             = _mm_add_ps(fix0,tx);
 304             fiy0             = _mm_add_ps(fiy0,ty);
 305             fiz0             = _mm_add_ps(fiz0,tz);
 306
 307             fjx0             = _mm_add_ps(fjx0,tx);
 308             fjy0             = _mm_add_ps(fjy0,ty);
 309             fjz0             = _mm_add_ps(fjz0,tz);
 310
 311             /**************************
 312              * CALCULATE INTERACTIONS *
 313              **************************/
 314
 315             if (gmx_mm_any_lt(rsq10,rcutoff2))
 316             {
 317
 318             /* Compute parameters for interactions between i and j atoms */
 319             qq10             = _mm_mul_ps(iq1,jq0);
 320
 321             /* REACTION-FIELD ELECTROSTATICS */
 322             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 323             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 324
 325             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 326
 327             /* Update potential sum for this i atom from the interaction with this j atom. */
 328             velec            = _mm_and_ps(velec,cutoff_mask);
 329             velecsum         = _mm_add_ps(velecsum,velec);
 330
 331             fscal            = felec;
 332
 333             fscal            = _mm_and_ps(fscal,cutoff_mask);
 334
 335             /* Calculate temporary vectorial force */
 336             tx               = _mm_mul_ps(fscal,dx10);
 337             ty               = _mm_mul_ps(fscal,dy10);
 338             tz               = _mm_mul_ps(fscal,dz10);
 339
 340             /* Update vectorial force */
 341             fix1             = _mm_add_ps(fix1,tx);
 342             fiy1             = _mm_add_ps(fiy1,ty);
 343             fiz1             = _mm_add_ps(fiz1,tz);
 344
 345             fjx0             = _mm_add_ps(fjx0,tx);
 346             fjy0             = _mm_add_ps(fjy0,ty);
 347             fjz0             = _mm_add_ps(fjz0,tz);
 348
 349             }
 350
 351             /**************************
 352              * CALCULATE INTERACTIONS *
 353              **************************/
 354
 355             if (gmx_mm_any_lt(rsq20,rcutoff2))
 356             {
 357
 358             /* Compute parameters for interactions between i and j atoms */
 359             qq20             = _mm_mul_ps(iq2,jq0);
 360
 361             /* REACTION-FIELD ELECTROSTATICS */
 362             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 363             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 364
 365             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 366
 367             /* Update potential sum for this i atom from the interaction with this j atom. */
 368             velec            = _mm_and_ps(velec,cutoff_mask);
 369             velecsum         = _mm_add_ps(velecsum,velec);
 370
 371             fscal            = felec;
 372
 373             fscal            = _mm_and_ps(fscal,cutoff_mask);
 374
 375             /* Calculate temporary vectorial force */
 376             tx               = _mm_mul_ps(fscal,dx20);
 377             ty               = _mm_mul_ps(fscal,dy20);
 378             tz               = _mm_mul_ps(fscal,dz20);
 379
 380             /* Update vectorial force */
 381             fix2             = _mm_add_ps(fix2,tx);
 382             fiy2             = _mm_add_ps(fiy2,ty);
 383             fiz2             = _mm_add_ps(fiz2,tz);
 384
 385             fjx0             = _mm_add_ps(fjx0,tx);
 386             fjy0             = _mm_add_ps(fjy0,ty);
 387             fjz0             = _mm_add_ps(fjz0,tz);
 388
 389             }
 390
 391             /**************************
 392              * CALCULATE INTERACTIONS *
 393              **************************/
 394
 395             if (gmx_mm_any_lt(rsq30,rcutoff2))
 396             {
 397
 398             /* Compute parameters for interactions between i and j atoms */
 399             qq30             = _mm_mul_ps(iq3,jq0);
 400
 401             /* REACTION-FIELD ELECTROSTATICS */
 402             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 403             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 404
 405             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 406
 407             /* Update potential sum for this i atom from the interaction with this j atom. */
 408             velec            = _mm_and_ps(velec,cutoff_mask);
 409             velecsum         = _mm_add_ps(velecsum,velec);
 410
 411             fscal            = felec;
 412
 413             fscal            = _mm_and_ps(fscal,cutoff_mask);
 414
 415             /* Calculate temporary vectorial force */
 416             tx               = _mm_mul_ps(fscal,dx30);
 417             ty               = _mm_mul_ps(fscal,dy30);
 418             tz               = _mm_mul_ps(fscal,dz30);
 419
 420             /* Update vectorial force */
 421             fix3             = _mm_add_ps(fix3,tx);
 422             fiy3             = _mm_add_ps(fiy3,ty);
 423             fiz3             = _mm_add_ps(fiz3,tz);
 424
 425             fjx0             = _mm_add_ps(fjx0,tx);
 426             fjy0             = _mm_add_ps(fjy0,ty);
 427             fjz0             = _mm_add_ps(fjz0,tz);
 428
 429             }
 430
 431             fjptrA             = f+j_coord_offsetA;
 432             fjptrB             = f+j_coord_offsetB;
 433             fjptrC             = f+j_coord_offsetC;
 434             fjptrD             = f+j_coord_offsetD;
 435
 436             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 437
 438             /* Inner loop uses 164 flops */
 439         }
 440
 441         if(jidx<j_index_end)
 442         {
 443
 444             /* Get j neighbor index, and coordinate index */
 445             jnrlistA         = jjnr[jidx];
 446             jnrlistB         = jjnr[jidx+1];
 447             jnrlistC         = jjnr[jidx+2];
 448             jnrlistD         = jjnr[jidx+3];
 449             /* Sign of each element will be negative for non-real atoms.
 450              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 451              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 452              */
 453             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 454             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 455             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 456             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 457             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 458             j_coord_offsetA  = DIM*jnrA;
 459             j_coord_offsetB  = DIM*jnrB;
 460             j_coord_offsetC  = DIM*jnrC;
 461             j_coord_offsetD  = DIM*jnrD;
 462
 463             /* load j atom coordinates */
 464             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 465                                               x+j_coord_offsetC,x+j_coord_offsetD,
 466                                               &jx0,&jy0,&jz0);
 467
 468             /* Calculate displacement vector */
 469             dx00             = _mm_sub_ps(ix0,jx0);
 470             dy00             = _mm_sub_ps(iy0,jy0);
 471             dz00             = _mm_sub_ps(iz0,jz0);
 472             dx10             = _mm_sub_ps(ix1,jx0);
 473             dy10             = _mm_sub_ps(iy1,jy0);
 474             dz10             = _mm_sub_ps(iz1,jz0);
 475             dx20             = _mm_sub_ps(ix2,jx0);
 476             dy20             = _mm_sub_ps(iy2,jy0);
 477             dz20             = _mm_sub_ps(iz2,jz0);
 478             dx30             = _mm_sub_ps(ix3,jx0);
 479             dy30             = _mm_sub_ps(iy3,jy0);
 480             dz30             = _mm_sub_ps(iz3,jz0);
 481
 482             /* Calculate squared distance and things based on it */
 483             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 484             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 485             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 486             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 487
 488             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 489             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 490             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 491             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 492
 493             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 494             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 495             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 496
 497             /* Load parameters for j particles */
 498             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 499                                                               charge+jnrC+0,charge+jnrD+0);
 500             vdwjidx0A        = 2*vdwtype[jnrA+0];
 501             vdwjidx0B        = 2*vdwtype[jnrB+0];
 502             vdwjidx0C        = 2*vdwtype[jnrC+0];
 503             vdwjidx0D        = 2*vdwtype[jnrD+0];
 504
 505             fjx0             = _mm_setzero_ps();
 506             fjy0             = _mm_setzero_ps();
 507             fjz0             = _mm_setzero_ps();
 508
 509             /**************************
 510              * CALCULATE INTERACTIONS *
 511              **************************/
 512
 513             r00              = _mm_mul_ps(rsq00,rinv00);
 514             r00              = _mm_andnot_ps(dummy_mask,r00);
 515
 516             /* Compute parameters for interactions between i and j atoms */
 517             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 518                                          vdwparam+vdwioffset0+vdwjidx0B,
 519                                          vdwparam+vdwioffset0+vdwjidx0C,
 520                                          vdwparam+vdwioffset0+vdwjidx0D,
 521                                          &c6_00,&c12_00);
 522
 523             /* Calculate table index by multiplying r with table scale and truncate to integer */
 524             rt               = _mm_mul_ps(r00,vftabscale);
 525             vfitab           = _mm_cvttps_epi32(rt);
 526             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 527             vfitab           = _mm_slli_epi32(vfitab,3);
 528
 529             /* CUBIC SPLINE TABLE DISPERSION */
 530             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 531             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 532             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 533             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 534             _MM_TRANSPOSE4_PS(Y,F,G,H);
 535             Heps             = _mm_mul_ps(vfeps,H);
 536             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 537             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 538             vvdw6            = _mm_mul_ps(c6_00,VV);
 539             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 540             fvdw6            = _mm_mul_ps(c6_00,FF);
 541
 542             /* CUBIC SPLINE TABLE REPULSION */
 543             vfitab           = _mm_add_epi32(vfitab,ifour);
 544             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 545             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 546             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 547             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 548             _MM_TRANSPOSE4_PS(Y,F,G,H);
 549             Heps             = _mm_mul_ps(vfeps,H);
 550             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 551             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 552             vvdw12           = _mm_mul_ps(c12_00,VV);
 553             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 554             fvdw12           = _mm_mul_ps(c12_00,FF);
 555             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 556             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 557
 558             /* Update potential sum for this i atom from the interaction with this j atom. */
 559             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 560             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 561
 562             fscal            = fvdw;
 563
 564             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 565
 566             /* Calculate temporary vectorial force */
 567             tx               = _mm_mul_ps(fscal,dx00);
 568             ty               = _mm_mul_ps(fscal,dy00);
 569             tz               = _mm_mul_ps(fscal,dz00);
 570
 571             /* Update vectorial force */
 572             fix0             = _mm_add_ps(fix0,tx);
 573             fiy0             = _mm_add_ps(fiy0,ty);
 574             fiz0             = _mm_add_ps(fiz0,tz);
 575
 576             fjx0             = _mm_add_ps(fjx0,tx);
 577             fjy0             = _mm_add_ps(fjy0,ty);
 578             fjz0             = _mm_add_ps(fjz0,tz);
 579
 580             /**************************
 581              * CALCULATE INTERACTIONS *
 582              **************************/
 583
 584             if (gmx_mm_any_lt(rsq10,rcutoff2))
 585             {
 586
 587             /* Compute parameters for interactions between i and j atoms */
 588             qq10             = _mm_mul_ps(iq1,jq0);
 589
 590             /* REACTION-FIELD ELECTROSTATICS */
 591             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 592             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 593
 594             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 595
 596             /* Update potential sum for this i atom from the interaction with this j atom. */
 597             velec            = _mm_and_ps(velec,cutoff_mask);
 598             velec            = _mm_andnot_ps(dummy_mask,velec);
 599             velecsum         = _mm_add_ps(velecsum,velec);
 600
 601             fscal            = felec;
 602
 603             fscal            = _mm_and_ps(fscal,cutoff_mask);
 604
 605             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 606
 607             /* Calculate temporary vectorial force */
 608             tx               = _mm_mul_ps(fscal,dx10);
 609             ty               = _mm_mul_ps(fscal,dy10);
 610             tz               = _mm_mul_ps(fscal,dz10);
 611
 612             /* Update vectorial force */
 613             fix1             = _mm_add_ps(fix1,tx);
 614             fiy1             = _mm_add_ps(fiy1,ty);
 615             fiz1             = _mm_add_ps(fiz1,tz);
 616
 617             fjx0             = _mm_add_ps(fjx0,tx);
 618             fjy0             = _mm_add_ps(fjy0,ty);
 619             fjz0             = _mm_add_ps(fjz0,tz);
 620
 621             }
 622
 623             /**************************
 624              * CALCULATE INTERACTIONS *
 625              **************************/
 626
 627             if (gmx_mm_any_lt(rsq20,rcutoff2))
 628             {
 629
 630             /* Compute parameters for interactions between i and j atoms */
 631             qq20             = _mm_mul_ps(iq2,jq0);
 632
 633             /* REACTION-FIELD ELECTROSTATICS */
 634             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 635             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 636
 637             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 638
 639             /* Update potential sum for this i atom from the interaction with this j atom. */
 640             velec            = _mm_and_ps(velec,cutoff_mask);
 641             velec            = _mm_andnot_ps(dummy_mask,velec);
 642             velecsum         = _mm_add_ps(velecsum,velec);
 643
 644             fscal            = felec;
 645
 646             fscal            = _mm_and_ps(fscal,cutoff_mask);
 647
 648             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 649
 650             /* Calculate temporary vectorial force */
 651             tx               = _mm_mul_ps(fscal,dx20);
 652             ty               = _mm_mul_ps(fscal,dy20);
 653             tz               = _mm_mul_ps(fscal,dz20);
 654
 655             /* Update vectorial force */
 656             fix2             = _mm_add_ps(fix2,tx);
 657             fiy2             = _mm_add_ps(fiy2,ty);
 658             fiz2             = _mm_add_ps(fiz2,tz);
 659
 660             fjx0             = _mm_add_ps(fjx0,tx);
 661             fjy0             = _mm_add_ps(fjy0,ty);
 662             fjz0             = _mm_add_ps(fjz0,tz);
 663
 664             }
 665
 666             /**************************
 667              * CALCULATE INTERACTIONS *
 668              **************************/
 669
 670             if (gmx_mm_any_lt(rsq30,rcutoff2))
 671             {
 672
 673             /* Compute parameters for interactions between i and j atoms */
 674             qq30             = _mm_mul_ps(iq3,jq0);
 675
 676             /* REACTION-FIELD ELECTROSTATICS */
 677             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 678             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 679
 680             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 681
 682             /* Update potential sum for this i atom from the interaction with this j atom. */
 683             velec            = _mm_and_ps(velec,cutoff_mask);
 684             velec            = _mm_andnot_ps(dummy_mask,velec);
 685             velecsum         = _mm_add_ps(velecsum,velec);
 686
 687             fscal            = felec;
 688
 689             fscal            = _mm_and_ps(fscal,cutoff_mask);
 690
 691             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 692
 693             /* Calculate temporary vectorial force */
 694             tx               = _mm_mul_ps(fscal,dx30);
 695             ty               = _mm_mul_ps(fscal,dy30);
 696             tz               = _mm_mul_ps(fscal,dz30);
 697
 698             /* Update vectorial force */
 699             fix3             = _mm_add_ps(fix3,tx);
 700             fiy3             = _mm_add_ps(fiy3,ty);
 701             fiz3             = _mm_add_ps(fiz3,tz);
 702
 703             fjx0             = _mm_add_ps(fjx0,tx);
 704             fjy0             = _mm_add_ps(fjy0,ty);
 705             fjz0             = _mm_add_ps(fjz0,tz);
 706
 707             }
 708
 709             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 710             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 711             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 712             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 713
 714             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 715
 716             /* Inner loop uses 165 flops */
 717         }
 718
 719         /* End of innermost loop */
 720
 721         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 722                                               f+i_coord_offset,fshift+i_shift_offset);
 723
 724         ggid                        = gid[iidx];
 725         /* Update potential energies */
 726         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 727         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 728
 729         /* Increment number of inner iterations */
 730         inneriter                  += j_index_end - j_index_start;
 731
 732         /* Outer loop uses 26 flops */
 733     }
 734
 735     /* Increment number of outer iterations */
 736     outeriter        += nri;
 737
 738     /* Update outer/inner flops */
 739
 740     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*165);
 741 }
 742 /*
 743  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_sse4_1_single
 744  * Electrostatics interaction: ReactionField
 745  * VdW interaction:            CubicSplineTable
 746  * Geometry:                   Water4-Particle
 747  * Calculate force/pot:        Force
 748  */
 749 void
 750 nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_sse4_1_single
 751                     (t_nblist * gmx_restrict                nlist,
 752                      rvec * gmx_restrict                    xx,
 753                      rvec * gmx_restrict                    ff,
 754                      t_forcerec * gmx_restrict              fr,
 755                      t_mdatoms * gmx_restrict               mdatoms,
 756                      nb_kernel_data_t * gmx_restrict        kernel_data,
 757                      t_nrnb * gmx_restrict                  nrnb)
 758 {
 759     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 760      * just 0 for non-waters.
 761      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 762      * jnr indices corresponding to data put in the four positions in the SIMD register.
 763      */
 764     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 765     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 766     int              jnrA,jnrB,jnrC,jnrD;
 767     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 768     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 769     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 770     real             rcutoff_scalar;
 771     real             *shiftvec,*fshift,*x,*f;
 772     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 773     real             scratch[4*DIM];
 774     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 775     int              vdwioffset0;
 776     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 777     int              vdwioffset1;
 778     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 779     int              vdwioffset2;
 780     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 781     int              vdwioffset3;
 782     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 783     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 784     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 785     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 786     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 787     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 788     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 789     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 790     real             *charge;
 791     int              nvdwtype;
 792     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 793     int              *vdwtype;
 794     real             *vdwparam;
 795     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 796     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 797     __m128i          vfitab;
 798     __m128i          ifour       = _mm_set1_epi32(4);
 799     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 800     real             *vftab;
 801     __m128           dummy_mask,cutoff_mask;
 802     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 803     __m128           one     = _mm_set1_ps(1.0);
 804     __m128           two     = _mm_set1_ps(2.0);
 805     x                = xx[0];
 806     f                = ff[0];
 807
 808     nri              = nlist->nri;
 809     iinr             = nlist->iinr;
 810     jindex           = nlist->jindex;
 811     jjnr             = nlist->jjnr;
 812     shiftidx         = nlist->shift;
 813     gid              = nlist->gid;
 814     shiftvec         = fr->shift_vec[0];
 815     fshift           = fr->fshift[0];
 816     facel            = _mm_set1_ps(fr->epsfac);
 817     charge           = mdatoms->chargeA;
 818     krf              = _mm_set1_ps(fr->ic->k_rf);
 819     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 820     crf              = _mm_set1_ps(fr->ic->c_rf);
 821     nvdwtype         = fr->ntype;
 822     vdwparam         = fr->nbfp;
 823     vdwtype          = mdatoms->typeA;
 824
 825     vftab            = kernel_data->table_vdw->data;
 826     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 827
 828     /* Setup water-specific parameters */
 829     inr              = nlist->iinr[0];
 830     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 831     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 832     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 833     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 834
 835     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 836     rcutoff_scalar   = fr->rcoulomb;
 837     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 838     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 839
 840     /* Avoid stupid compiler warnings */
 841     jnrA = jnrB = jnrC = jnrD = 0;
 842     j_coord_offsetA = 0;
 843     j_coord_offsetB = 0;
 844     j_coord_offsetC = 0;
 845     j_coord_offsetD = 0;
 846
 847     outeriter        = 0;
 848     inneriter        = 0;
 849
 850     for(iidx=0;iidx<4*DIM;iidx++)
 851     {
 852         scratch[iidx] = 0.0;
 853     }
 854
 855     /* Start outer loop over neighborlists */
 856     for(iidx=0; iidx<nri; iidx++)
 857     {
 858         /* Load shift vector for this list */
 859         i_shift_offset   = DIM*shiftidx[iidx];
 860
 861         /* Load limits for loop over neighbors */
 862         j_index_start    = jindex[iidx];
 863         j_index_end      = jindex[iidx+1];
 864
 865         /* Get outer coordinate index */
 866         inr              = iinr[iidx];
 867         i_coord_offset   = DIM*inr;
 868
 869         /* Load i particle coords and add shift vector */
 870         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 871                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 872
 873         fix0             = _mm_setzero_ps();
 874         fiy0             = _mm_setzero_ps();
 875         fiz0             = _mm_setzero_ps();
 876         fix1             = _mm_setzero_ps();
 877         fiy1             = _mm_setzero_ps();
 878         fiz1             = _mm_setzero_ps();
 879         fix2             = _mm_setzero_ps();
 880         fiy2             = _mm_setzero_ps();
 881         fiz2             = _mm_setzero_ps();
 882         fix3             = _mm_setzero_ps();
 883         fiy3             = _mm_setzero_ps();
 884         fiz3             = _mm_setzero_ps();
 885
 886         /* Start inner kernel loop */
 887         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 888         {
 889
 890             /* Get j neighbor index, and coordinate index */
 891             jnrA             = jjnr[jidx];
 892             jnrB             = jjnr[jidx+1];
 893             jnrC             = jjnr[jidx+2];
 894             jnrD             = jjnr[jidx+3];
 895             j_coord_offsetA  = DIM*jnrA;
 896             j_coord_offsetB  = DIM*jnrB;
 897             j_coord_offsetC  = DIM*jnrC;
 898             j_coord_offsetD  = DIM*jnrD;
 899
 900             /* load j atom coordinates */
 901             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 902                                               x+j_coord_offsetC,x+j_coord_offsetD,
 903                                               &jx0,&jy0,&jz0);
 904
 905             /* Calculate displacement vector */
 906             dx00             = _mm_sub_ps(ix0,jx0);
 907             dy00             = _mm_sub_ps(iy0,jy0);
 908             dz00             = _mm_sub_ps(iz0,jz0);
 909             dx10             = _mm_sub_ps(ix1,jx0);
 910             dy10             = _mm_sub_ps(iy1,jy0);
 911             dz10             = _mm_sub_ps(iz1,jz0);
 912             dx20             = _mm_sub_ps(ix2,jx0);
 913             dy20             = _mm_sub_ps(iy2,jy0);
 914             dz20             = _mm_sub_ps(iz2,jz0);
 915             dx30             = _mm_sub_ps(ix3,jx0);
 916             dy30             = _mm_sub_ps(iy3,jy0);
 917             dz30             = _mm_sub_ps(iz3,jz0);
 918
 919             /* Calculate squared distance and things based on it */
 920             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 921             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 922             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 923             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 924
 925             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 926             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 927             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 928             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 929
 930             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 931             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 932             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 933
 934             /* Load parameters for j particles */
 935             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 936                                                               charge+jnrC+0,charge+jnrD+0);
 937             vdwjidx0A        = 2*vdwtype[jnrA+0];
 938             vdwjidx0B        = 2*vdwtype[jnrB+0];
 939             vdwjidx0C        = 2*vdwtype[jnrC+0];
 940             vdwjidx0D        = 2*vdwtype[jnrD+0];
 941
 942             fjx0             = _mm_setzero_ps();
 943             fjy0             = _mm_setzero_ps();
 944             fjz0             = _mm_setzero_ps();
 945
 946             /**************************
 947              * CALCULATE INTERACTIONS *
 948              **************************/
 949
 950             r00              = _mm_mul_ps(rsq00,rinv00);
 951
 952             /* Compute parameters for interactions between i and j atoms */
 953             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 954                                          vdwparam+vdwioffset0+vdwjidx0B,
 955                                          vdwparam+vdwioffset0+vdwjidx0C,
 956                                          vdwparam+vdwioffset0+vdwjidx0D,
 957                                          &c6_00,&c12_00);
 958
 959             /* Calculate table index by multiplying r with table scale and truncate to integer */
 960             rt               = _mm_mul_ps(r00,vftabscale);
 961             vfitab           = _mm_cvttps_epi32(rt);
 962             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 963             vfitab           = _mm_slli_epi32(vfitab,3);
 964
 965             /* CUBIC SPLINE TABLE DISPERSION */
 966             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 967             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 968             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 969             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 970             _MM_TRANSPOSE4_PS(Y,F,G,H);
 971             Heps             = _mm_mul_ps(vfeps,H);
 972             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 973             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 974             fvdw6            = _mm_mul_ps(c6_00,FF);
 975
 976             /* CUBIC SPLINE TABLE REPULSION */
 977             vfitab           = _mm_add_epi32(vfitab,ifour);
 978             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 979             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 980             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 981             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 982             _MM_TRANSPOSE4_PS(Y,F,G,H);
 983             Heps             = _mm_mul_ps(vfeps,H);
 984             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 985             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 986             fvdw12           = _mm_mul_ps(c12_00,FF);
 987             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 988
 989             fscal            = fvdw;
 990
 991             /* Calculate temporary vectorial force */
 992             tx               = _mm_mul_ps(fscal,dx00);
 993             ty               = _mm_mul_ps(fscal,dy00);
 994             tz               = _mm_mul_ps(fscal,dz00);
 995
 996             /* Update vectorial force */
 997             fix0             = _mm_add_ps(fix0,tx);
 998             fiy0             = _mm_add_ps(fiy0,ty);
 999             fiz0             = _mm_add_ps(fiz0,tz);
1000
1001             fjx0             = _mm_add_ps(fjx0,tx);
1002             fjy0             = _mm_add_ps(fjy0,ty);
1003             fjz0             = _mm_add_ps(fjz0,tz);
1004
1005             /**************************
1006              * CALCULATE INTERACTIONS *
1007              **************************/
1008
1009             if (gmx_mm_any_lt(rsq10,rcutoff2))
1010             {
1011
1012             /* Compute parameters for interactions between i and j atoms */
1013             qq10             = _mm_mul_ps(iq1,jq0);
1014
1015             /* REACTION-FIELD ELECTROSTATICS */
1016             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1017
1018             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1019
1020             fscal            = felec;
1021
1022             fscal            = _mm_and_ps(fscal,cutoff_mask);
1023
1024             /* Calculate temporary vectorial force */
1025             tx               = _mm_mul_ps(fscal,dx10);
1026             ty               = _mm_mul_ps(fscal,dy10);
1027             tz               = _mm_mul_ps(fscal,dz10);
1028
1029             /* Update vectorial force */
1030             fix1             = _mm_add_ps(fix1,tx);
1031             fiy1             = _mm_add_ps(fiy1,ty);
1032             fiz1             = _mm_add_ps(fiz1,tz);
1033
1034             fjx0             = _mm_add_ps(fjx0,tx);
1035             fjy0             = _mm_add_ps(fjy0,ty);
1036             fjz0             = _mm_add_ps(fjz0,tz);
1037
1038             }
1039
1040             /**************************
1041              * CALCULATE INTERACTIONS *
1042              **************************/
1043
1044             if (gmx_mm_any_lt(rsq20,rcutoff2))
1045             {
1046
1047             /* Compute parameters for interactions between i and j atoms */
1048             qq20             = _mm_mul_ps(iq2,jq0);
1049
1050             /* REACTION-FIELD ELECTROSTATICS */
1051             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1052
1053             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1054
1055             fscal            = felec;
1056
1057             fscal            = _mm_and_ps(fscal,cutoff_mask);
1058
1059             /* Calculate temporary vectorial force */
1060             tx               = _mm_mul_ps(fscal,dx20);
1061             ty               = _mm_mul_ps(fscal,dy20);
1062             tz               = _mm_mul_ps(fscal,dz20);
1063
1064             /* Update vectorial force */
1065             fix2             = _mm_add_ps(fix2,tx);
1066             fiy2             = _mm_add_ps(fiy2,ty);
1067             fiz2             = _mm_add_ps(fiz2,tz);
1068
1069             fjx0             = _mm_add_ps(fjx0,tx);
1070             fjy0             = _mm_add_ps(fjy0,ty);
1071             fjz0             = _mm_add_ps(fjz0,tz);
1072
1073             }
1074
1075             /**************************
1076              * CALCULATE INTERACTIONS *
1077              **************************/
1078
1079             if (gmx_mm_any_lt(rsq30,rcutoff2))
1080             {
1081
1082             /* Compute parameters for interactions between i and j atoms */
1083             qq30             = _mm_mul_ps(iq3,jq0);
1084
1085             /* REACTION-FIELD ELECTROSTATICS */
1086             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1087
1088             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1089
1090             fscal            = felec;
1091
1092             fscal            = _mm_and_ps(fscal,cutoff_mask);
1093
1094             /* Calculate temporary vectorial force */
1095             tx               = _mm_mul_ps(fscal,dx30);
1096             ty               = _mm_mul_ps(fscal,dy30);
1097             tz               = _mm_mul_ps(fscal,dz30);
1098
1099             /* Update vectorial force */
1100             fix3             = _mm_add_ps(fix3,tx);
1101             fiy3             = _mm_add_ps(fiy3,ty);
1102             fiz3             = _mm_add_ps(fiz3,tz);
1103
1104             fjx0             = _mm_add_ps(fjx0,tx);
1105             fjy0             = _mm_add_ps(fjy0,ty);
1106             fjz0             = _mm_add_ps(fjz0,tz);
1107
1108             }
1109
1110             fjptrA             = f+j_coord_offsetA;
1111             fjptrB             = f+j_coord_offsetB;
1112             fjptrC             = f+j_coord_offsetC;
1113             fjptrD             = f+j_coord_offsetD;
1114
1115             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1116
1117             /* Inner loop uses 138 flops */
1118         }
1119
1120         if(jidx<j_index_end)
1121         {
1122
1123             /* Get j neighbor index, and coordinate index */
1124             jnrlistA         = jjnr[jidx];
1125             jnrlistB         = jjnr[jidx+1];
1126             jnrlistC         = jjnr[jidx+2];
1127             jnrlistD         = jjnr[jidx+3];
1128             /* Sign of each element will be negative for non-real atoms.
1129              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1130              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1131              */
1132             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1133             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1134             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1135             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1136             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1137             j_coord_offsetA  = DIM*jnrA;
1138             j_coord_offsetB  = DIM*jnrB;
1139             j_coord_offsetC  = DIM*jnrC;
1140             j_coord_offsetD  = DIM*jnrD;
1141
1142             /* load j atom coordinates */
1143             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1144                                               x+j_coord_offsetC,x+j_coord_offsetD,
1145                                               &jx0,&jy0,&jz0);
1146
1147             /* Calculate displacement vector */
1148             dx00             = _mm_sub_ps(ix0,jx0);
1149             dy00             = _mm_sub_ps(iy0,jy0);
1150             dz00             = _mm_sub_ps(iz0,jz0);
1151             dx10             = _mm_sub_ps(ix1,jx0);
1152             dy10             = _mm_sub_ps(iy1,jy0);
1153             dz10             = _mm_sub_ps(iz1,jz0);
1154             dx20             = _mm_sub_ps(ix2,jx0);
1155             dy20             = _mm_sub_ps(iy2,jy0);
1156             dz20             = _mm_sub_ps(iz2,jz0);
1157             dx30             = _mm_sub_ps(ix3,jx0);
1158             dy30             = _mm_sub_ps(iy3,jy0);
1159             dz30             = _mm_sub_ps(iz3,jz0);
1160
1161             /* Calculate squared distance and things based on it */
1162             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1163             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1164             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1165             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1166
1167             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1168             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1169             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1170             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1171
1172             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1173             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1174             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1175
1176             /* Load parameters for j particles */
1177             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1178                                                               charge+jnrC+0,charge+jnrD+0);
1179             vdwjidx0A        = 2*vdwtype[jnrA+0];
1180             vdwjidx0B        = 2*vdwtype[jnrB+0];
1181             vdwjidx0C        = 2*vdwtype[jnrC+0];
1182             vdwjidx0D        = 2*vdwtype[jnrD+0];
1183
1184             fjx0             = _mm_setzero_ps();
1185             fjy0             = _mm_setzero_ps();
1186             fjz0             = _mm_setzero_ps();
1187
1188             /**************************
1189              * CALCULATE INTERACTIONS *
1190              **************************/
1191
1192             r00              = _mm_mul_ps(rsq00,rinv00);
1193             r00              = _mm_andnot_ps(dummy_mask,r00);
1194
1195             /* Compute parameters for interactions between i and j atoms */
1196             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1197                                          vdwparam+vdwioffset0+vdwjidx0B,
1198                                          vdwparam+vdwioffset0+vdwjidx0C,
1199                                          vdwparam+vdwioffset0+vdwjidx0D,
1200                                          &c6_00,&c12_00);
1201
1202             /* Calculate table index by multiplying r with table scale and truncate to integer */
1203             rt               = _mm_mul_ps(r00,vftabscale);
1204             vfitab           = _mm_cvttps_epi32(rt);
1205             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1206             vfitab           = _mm_slli_epi32(vfitab,3);
1207
1208             /* CUBIC SPLINE TABLE DISPERSION */
1209             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1210             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1211             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1212             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1213             _MM_TRANSPOSE4_PS(Y,F,G,H);
1214             Heps             = _mm_mul_ps(vfeps,H);
1215             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1216             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1217             fvdw6            = _mm_mul_ps(c6_00,FF);
1218
1219             /* CUBIC SPLINE TABLE REPULSION */
1220             vfitab           = _mm_add_epi32(vfitab,ifour);
1221             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1222             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1223             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1224             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1225             _MM_TRANSPOSE4_PS(Y,F,G,H);
1226             Heps             = _mm_mul_ps(vfeps,H);
1227             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1228             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1229             fvdw12           = _mm_mul_ps(c12_00,FF);
1230             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1231
1232             fscal            = fvdw;
1233
1234             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1235
1236             /* Calculate temporary vectorial force */
1237             tx               = _mm_mul_ps(fscal,dx00);
1238             ty               = _mm_mul_ps(fscal,dy00);
1239             tz               = _mm_mul_ps(fscal,dz00);
1240
1241             /* Update vectorial force */
1242             fix0             = _mm_add_ps(fix0,tx);
1243             fiy0             = _mm_add_ps(fiy0,ty);
1244             fiz0             = _mm_add_ps(fiz0,tz);
1245
1246             fjx0             = _mm_add_ps(fjx0,tx);
1247             fjy0             = _mm_add_ps(fjy0,ty);
1248             fjz0             = _mm_add_ps(fjz0,tz);
1249
1250             /**************************
1251              * CALCULATE INTERACTIONS *
1252              **************************/
1253
1254             if (gmx_mm_any_lt(rsq10,rcutoff2))
1255             {
1256
1257             /* Compute parameters for interactions between i and j atoms */
1258             qq10             = _mm_mul_ps(iq1,jq0);
1259
1260             /* REACTION-FIELD ELECTROSTATICS */
1261             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1262
1263             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1264
1265             fscal            = felec;
1266
1267             fscal            = _mm_and_ps(fscal,cutoff_mask);
1268
1269             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1270
1271             /* Calculate temporary vectorial force */
1272             tx               = _mm_mul_ps(fscal,dx10);
1273             ty               = _mm_mul_ps(fscal,dy10);
1274             tz               = _mm_mul_ps(fscal,dz10);
1275
1276             /* Update vectorial force */
1277             fix1             = _mm_add_ps(fix1,tx);
1278             fiy1             = _mm_add_ps(fiy1,ty);
1279             fiz1             = _mm_add_ps(fiz1,tz);
1280
1281             fjx0             = _mm_add_ps(fjx0,tx);
1282             fjy0             = _mm_add_ps(fjy0,ty);
1283             fjz0             = _mm_add_ps(fjz0,tz);
1284
1285             }
1286
1287             /**************************
1288              * CALCULATE INTERACTIONS *
1289              **************************/
1290
1291             if (gmx_mm_any_lt(rsq20,rcutoff2))
1292             {
1293
1294             /* Compute parameters for interactions between i and j atoms */
1295             qq20             = _mm_mul_ps(iq2,jq0);
1296
1297             /* REACTION-FIELD ELECTROSTATICS */
1298             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1299
1300             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1301
1302             fscal            = felec;
1303
1304             fscal            = _mm_and_ps(fscal,cutoff_mask);
1305
1306             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1307
1308             /* Calculate temporary vectorial force */
1309             tx               = _mm_mul_ps(fscal,dx20);
1310             ty               = _mm_mul_ps(fscal,dy20);
1311             tz               = _mm_mul_ps(fscal,dz20);
1312
1313             /* Update vectorial force */
1314             fix2             = _mm_add_ps(fix2,tx);
1315             fiy2             = _mm_add_ps(fiy2,ty);
1316             fiz2             = _mm_add_ps(fiz2,tz);
1317
1318             fjx0             = _mm_add_ps(fjx0,tx);
1319             fjy0             = _mm_add_ps(fjy0,ty);
1320             fjz0             = _mm_add_ps(fjz0,tz);
1321
1322             }
1323
1324             /**************************
1325              * CALCULATE INTERACTIONS *
1326              **************************/
1327
1328             if (gmx_mm_any_lt(rsq30,rcutoff2))
1329             {
1330
1331             /* Compute parameters for interactions between i and j atoms */
1332             qq30             = _mm_mul_ps(iq3,jq0);
1333
1334             /* REACTION-FIELD ELECTROSTATICS */
1335             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1336
1337             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1338
1339             fscal            = felec;
1340
1341             fscal            = _mm_and_ps(fscal,cutoff_mask);
1342
1343             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1344
1345             /* Calculate temporary vectorial force */
1346             tx               = _mm_mul_ps(fscal,dx30);
1347             ty               = _mm_mul_ps(fscal,dy30);
1348             tz               = _mm_mul_ps(fscal,dz30);
1349
1350             /* Update vectorial force */
1351             fix3             = _mm_add_ps(fix3,tx);
1352             fiy3             = _mm_add_ps(fiy3,ty);
1353             fiz3             = _mm_add_ps(fiz3,tz);
1354
1355             fjx0             = _mm_add_ps(fjx0,tx);
1356             fjy0             = _mm_add_ps(fjy0,ty);
1357             fjz0             = _mm_add_ps(fjz0,tz);
1358
1359             }
1360
1361             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1362             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1363             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1364             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1365
1366             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1367
1368             /* Inner loop uses 139 flops */
1369         }
1370
1371         /* End of innermost loop */
1372
1373         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1374                                               f+i_coord_offset,fshift+i_shift_offset);
1375
1376         /* Increment number of inner iterations */
1377         inneriter                  += j_index_end - j_index_start;
1378
1379         /* Outer loop uses 24 flops */
1380     }
1381
1382     /* Increment number of outer iterations */
1383     outeriter        += nri;
1384
1385     /* Update outer/inner flops */
1386
1387     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*139);
1388 }