src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEw_VdwLJ_GeomW4P1_sse2_single.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse2_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW4P1_VF_sse2_single
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwLJ_GeomW4P1_VF_sse2_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwioffset3;
  89     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  91     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  96     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 103     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 104     __m128i          ewitab;
 105     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 106     real             *ewtab;
 107     __m128           dummy_mask,cutoff_mask;
 108     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 109     __m128           one     = _mm_set1_ps(1.0);
 110     __m128           two     = _mm_set1_ps(2.0);
 111     x                = xx[0];
 112     f                = ff[0];
 113
 114     nri              = nlist->nri;
 115     iinr             = nlist->iinr;
 116     jindex           = nlist->jindex;
 117     jjnr             = nlist->jjnr;
 118     shiftidx         = nlist->shift;
 119     gid              = nlist->gid;
 120     shiftvec         = fr->shift_vec[0];
 121     fshift           = fr->fshift[0];
 122     facel            = _mm_set1_ps(fr->ic->epsfac);
 123     charge           = mdatoms->chargeA;
 124     nvdwtype         = fr->ntype;
 125     vdwparam         = fr->nbfp;
 126     vdwtype          = mdatoms->typeA;
 127
 128     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 129     ewtab            = fr->ic->tabq_coul_FDV0;
 130     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 131     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 132
 133     /* Setup water-specific parameters */
 134     inr              = nlist->iinr[0];
 135     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 136     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 137     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 138     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 139
 140     /* Avoid stupid compiler warnings */
 141     jnrA = jnrB = jnrC = jnrD = 0;
 142     j_coord_offsetA = 0;
 143     j_coord_offsetB = 0;
 144     j_coord_offsetC = 0;
 145     j_coord_offsetD = 0;
 146
 147     outeriter        = 0;
 148     inneriter        = 0;
 149
 150     for(iidx=0;iidx<4*DIM;iidx++)
 151     {
 152         scratch[iidx] = 0.0;
 153     }
 154
 155     /* Start outer loop over neighborlists */
 156     for(iidx=0; iidx<nri; iidx++)
 157     {
 158         /* Load shift vector for this list */
 159         i_shift_offset   = DIM*shiftidx[iidx];
 160
 161         /* Load limits for loop over neighbors */
 162         j_index_start    = jindex[iidx];
 163         j_index_end      = jindex[iidx+1];
 164
 165         /* Get outer coordinate index */
 166         inr              = iinr[iidx];
 167         i_coord_offset   = DIM*inr;
 168
 169         /* Load i particle coords and add shift vector */
 170         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 171                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 172
 173         fix0             = _mm_setzero_ps();
 174         fiy0             = _mm_setzero_ps();
 175         fiz0             = _mm_setzero_ps();
 176         fix1             = _mm_setzero_ps();
 177         fiy1             = _mm_setzero_ps();
 178         fiz1             = _mm_setzero_ps();
 179         fix2             = _mm_setzero_ps();
 180         fiy2             = _mm_setzero_ps();
 181         fiz2             = _mm_setzero_ps();
 182         fix3             = _mm_setzero_ps();
 183         fiy3             = _mm_setzero_ps();
 184         fiz3             = _mm_setzero_ps();
 185
 186         /* Reset potential sums */
 187         velecsum         = _mm_setzero_ps();
 188         vvdwsum          = _mm_setzero_ps();
 189
 190         /* Start inner kernel loop */
 191         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 192         {
 193
 194             /* Get j neighbor index, and coordinate index */
 195             jnrA             = jjnr[jidx];
 196             jnrB             = jjnr[jidx+1];
 197             jnrC             = jjnr[jidx+2];
 198             jnrD             = jjnr[jidx+3];
 199             j_coord_offsetA  = DIM*jnrA;
 200             j_coord_offsetB  = DIM*jnrB;
 201             j_coord_offsetC  = DIM*jnrC;
 202             j_coord_offsetD  = DIM*jnrD;
 203
 204             /* load j atom coordinates */
 205             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 206                                               x+j_coord_offsetC,x+j_coord_offsetD,
 207                                               &jx0,&jy0,&jz0);
 208
 209             /* Calculate displacement vector */
 210             dx00             = _mm_sub_ps(ix0,jx0);
 211             dy00             = _mm_sub_ps(iy0,jy0);
 212             dz00             = _mm_sub_ps(iz0,jz0);
 213             dx10             = _mm_sub_ps(ix1,jx0);
 214             dy10             = _mm_sub_ps(iy1,jy0);
 215             dz10             = _mm_sub_ps(iz1,jz0);
 216             dx20             = _mm_sub_ps(ix2,jx0);
 217             dy20             = _mm_sub_ps(iy2,jy0);
 218             dz20             = _mm_sub_ps(iz2,jz0);
 219             dx30             = _mm_sub_ps(ix3,jx0);
 220             dy30             = _mm_sub_ps(iy3,jy0);
 221             dz30             = _mm_sub_ps(iz3,jz0);
 222
 223             /* Calculate squared distance and things based on it */
 224             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 225             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 226             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 227             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 228
 229             rinv10           = sse2_invsqrt_f(rsq10);
 230             rinv20           = sse2_invsqrt_f(rsq20);
 231             rinv30           = sse2_invsqrt_f(rsq30);
 232
 233             rinvsq00         = sse2_inv_f(rsq00);
 234             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 235             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 236             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 237
 238             /* Load parameters for j particles */
 239             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 240                                                               charge+jnrC+0,charge+jnrD+0);
 241             vdwjidx0A        = 2*vdwtype[jnrA+0];
 242             vdwjidx0B        = 2*vdwtype[jnrB+0];
 243             vdwjidx0C        = 2*vdwtype[jnrC+0];
 244             vdwjidx0D        = 2*vdwtype[jnrD+0];
 245
 246             fjx0             = _mm_setzero_ps();
 247             fjy0             = _mm_setzero_ps();
 248             fjz0             = _mm_setzero_ps();
 249
 250             /**************************
 251              * CALCULATE INTERACTIONS *
 252              **************************/
 253
 254             /* Compute parameters for interactions between i and j atoms */
 255             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 256                                          vdwparam+vdwioffset0+vdwjidx0B,
 257                                          vdwparam+vdwioffset0+vdwjidx0C,
 258                                          vdwparam+vdwioffset0+vdwjidx0D,
 259                                          &c6_00,&c12_00);
 260
 261             /* LENNARD-JONES DISPERSION/REPULSION */
 262
 263             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 264             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 265             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 266             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 267             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 268
 269             /* Update potential sum for this i atom from the interaction with this j atom. */
 270             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 271
 272             fscal            = fvdw;
 273
 274             /* Calculate temporary vectorial force */
 275             tx               = _mm_mul_ps(fscal,dx00);
 276             ty               = _mm_mul_ps(fscal,dy00);
 277             tz               = _mm_mul_ps(fscal,dz00);
 278
 279             /* Update vectorial force */
 280             fix0             = _mm_add_ps(fix0,tx);
 281             fiy0             = _mm_add_ps(fiy0,ty);
 282             fiz0             = _mm_add_ps(fiz0,tz);
 283
 284             fjx0             = _mm_add_ps(fjx0,tx);
 285             fjy0             = _mm_add_ps(fjy0,ty);
 286             fjz0             = _mm_add_ps(fjz0,tz);
 287
 288             /**************************
 289              * CALCULATE INTERACTIONS *
 290              **************************/
 291
 292             r10              = _mm_mul_ps(rsq10,rinv10);
 293
 294             /* Compute parameters for interactions between i and j atoms */
 295             qq10             = _mm_mul_ps(iq1,jq0);
 296
 297             /* EWALD ELECTROSTATICS */
 298
 299             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 300             ewrt             = _mm_mul_ps(r10,ewtabscale);
 301             ewitab           = _mm_cvttps_epi32(ewrt);
 302             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 303             ewitab           = _mm_slli_epi32(ewitab,2);
 304             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 305             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 306             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 307             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 308             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 309             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 310             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 311             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 312             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 313
 314             /* Update potential sum for this i atom from the interaction with this j atom. */
 315             velecsum         = _mm_add_ps(velecsum,velec);
 316
 317             fscal            = felec;
 318
 319             /* Calculate temporary vectorial force */
 320             tx               = _mm_mul_ps(fscal,dx10);
 321             ty               = _mm_mul_ps(fscal,dy10);
 322             tz               = _mm_mul_ps(fscal,dz10);
 323
 324             /* Update vectorial force */
 325             fix1             = _mm_add_ps(fix1,tx);
 326             fiy1             = _mm_add_ps(fiy1,ty);
 327             fiz1             = _mm_add_ps(fiz1,tz);
 328
 329             fjx0             = _mm_add_ps(fjx0,tx);
 330             fjy0             = _mm_add_ps(fjy0,ty);
 331             fjz0             = _mm_add_ps(fjz0,tz);
 332
 333             /**************************
 334              * CALCULATE INTERACTIONS *
 335              **************************/
 336
 337             r20              = _mm_mul_ps(rsq20,rinv20);
 338
 339             /* Compute parameters for interactions between i and j atoms */
 340             qq20             = _mm_mul_ps(iq2,jq0);
 341
 342             /* EWALD ELECTROSTATICS */
 343
 344             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 345             ewrt             = _mm_mul_ps(r20,ewtabscale);
 346             ewitab           = _mm_cvttps_epi32(ewrt);
 347             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 348             ewitab           = _mm_slli_epi32(ewitab,2);
 349             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 350             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 351             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 352             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 353             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 354             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 355             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 356             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 357             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 358
 359             /* Update potential sum for this i atom from the interaction with this j atom. */
 360             velecsum         = _mm_add_ps(velecsum,velec);
 361
 362             fscal            = felec;
 363
 364             /* Calculate temporary vectorial force */
 365             tx               = _mm_mul_ps(fscal,dx20);
 366             ty               = _mm_mul_ps(fscal,dy20);
 367             tz               = _mm_mul_ps(fscal,dz20);
 368
 369             /* Update vectorial force */
 370             fix2             = _mm_add_ps(fix2,tx);
 371             fiy2             = _mm_add_ps(fiy2,ty);
 372             fiz2             = _mm_add_ps(fiz2,tz);
 373
 374             fjx0             = _mm_add_ps(fjx0,tx);
 375             fjy0             = _mm_add_ps(fjy0,ty);
 376             fjz0             = _mm_add_ps(fjz0,tz);
 377
 378             /**************************
 379              * CALCULATE INTERACTIONS *
 380              **************************/
 381
 382             r30              = _mm_mul_ps(rsq30,rinv30);
 383
 384             /* Compute parameters for interactions between i and j atoms */
 385             qq30             = _mm_mul_ps(iq3,jq0);
 386
 387             /* EWALD ELECTROSTATICS */
 388
 389             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 390             ewrt             = _mm_mul_ps(r30,ewtabscale);
 391             ewitab           = _mm_cvttps_epi32(ewrt);
 392             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 393             ewitab           = _mm_slli_epi32(ewitab,2);
 394             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 395             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 396             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 397             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 398             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 399             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 400             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 401             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 402             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 403
 404             /* Update potential sum for this i atom from the interaction with this j atom. */
 405             velecsum         = _mm_add_ps(velecsum,velec);
 406
 407             fscal            = felec;
 408
 409             /* Calculate temporary vectorial force */
 410             tx               = _mm_mul_ps(fscal,dx30);
 411             ty               = _mm_mul_ps(fscal,dy30);
 412             tz               = _mm_mul_ps(fscal,dz30);
 413
 414             /* Update vectorial force */
 415             fix3             = _mm_add_ps(fix3,tx);
 416             fiy3             = _mm_add_ps(fiy3,ty);
 417             fiz3             = _mm_add_ps(fiz3,tz);
 418
 419             fjx0             = _mm_add_ps(fjx0,tx);
 420             fjy0             = _mm_add_ps(fjy0,ty);
 421             fjz0             = _mm_add_ps(fjz0,tz);
 422
 423             fjptrA             = f+j_coord_offsetA;
 424             fjptrB             = f+j_coord_offsetB;
 425             fjptrC             = f+j_coord_offsetC;
 426             fjptrD             = f+j_coord_offsetD;
 427
 428             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 429
 430             /* Inner loop uses 155 flops */
 431         }
 432
 433         if(jidx<j_index_end)
 434         {
 435
 436             /* Get j neighbor index, and coordinate index */
 437             jnrlistA         = jjnr[jidx];
 438             jnrlistB         = jjnr[jidx+1];
 439             jnrlistC         = jjnr[jidx+2];
 440             jnrlistD         = jjnr[jidx+3];
 441             /* Sign of each element will be negative for non-real atoms.
 442              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 443              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 444              */
 445             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 446             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 447             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 448             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 449             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 450             j_coord_offsetA  = DIM*jnrA;
 451             j_coord_offsetB  = DIM*jnrB;
 452             j_coord_offsetC  = DIM*jnrC;
 453             j_coord_offsetD  = DIM*jnrD;
 454
 455             /* load j atom coordinates */
 456             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 457                                               x+j_coord_offsetC,x+j_coord_offsetD,
 458                                               &jx0,&jy0,&jz0);
 459
 460             /* Calculate displacement vector */
 461             dx00             = _mm_sub_ps(ix0,jx0);
 462             dy00             = _mm_sub_ps(iy0,jy0);
 463             dz00             = _mm_sub_ps(iz0,jz0);
 464             dx10             = _mm_sub_ps(ix1,jx0);
 465             dy10             = _mm_sub_ps(iy1,jy0);
 466             dz10             = _mm_sub_ps(iz1,jz0);
 467             dx20             = _mm_sub_ps(ix2,jx0);
 468             dy20             = _mm_sub_ps(iy2,jy0);
 469             dz20             = _mm_sub_ps(iz2,jz0);
 470             dx30             = _mm_sub_ps(ix3,jx0);
 471             dy30             = _mm_sub_ps(iy3,jy0);
 472             dz30             = _mm_sub_ps(iz3,jz0);
 473
 474             /* Calculate squared distance and things based on it */
 475             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 476             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 477             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 478             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 479
 480             rinv10           = sse2_invsqrt_f(rsq10);
 481             rinv20           = sse2_invsqrt_f(rsq20);
 482             rinv30           = sse2_invsqrt_f(rsq30);
 483
 484             rinvsq00         = sse2_inv_f(rsq00);
 485             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 486             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 487             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 488
 489             /* Load parameters for j particles */
 490             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 491                                                               charge+jnrC+0,charge+jnrD+0);
 492             vdwjidx0A        = 2*vdwtype[jnrA+0];
 493             vdwjidx0B        = 2*vdwtype[jnrB+0];
 494             vdwjidx0C        = 2*vdwtype[jnrC+0];
 495             vdwjidx0D        = 2*vdwtype[jnrD+0];
 496
 497             fjx0             = _mm_setzero_ps();
 498             fjy0             = _mm_setzero_ps();
 499             fjz0             = _mm_setzero_ps();
 500
 501             /**************************
 502              * CALCULATE INTERACTIONS *
 503              **************************/
 504
 505             /* Compute parameters for interactions between i and j atoms */
 506             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 507                                          vdwparam+vdwioffset0+vdwjidx0B,
 508                                          vdwparam+vdwioffset0+vdwjidx0C,
 509                                          vdwparam+vdwioffset0+vdwjidx0D,
 510                                          &c6_00,&c12_00);
 511
 512             /* LENNARD-JONES DISPERSION/REPULSION */
 513
 514             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 515             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 516             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 517             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 518             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 519
 520             /* Update potential sum for this i atom from the interaction with this j atom. */
 521             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 522             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 523
 524             fscal            = fvdw;
 525
 526             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 527
 528             /* Calculate temporary vectorial force */
 529             tx               = _mm_mul_ps(fscal,dx00);
 530             ty               = _mm_mul_ps(fscal,dy00);
 531             tz               = _mm_mul_ps(fscal,dz00);
 532
 533             /* Update vectorial force */
 534             fix0             = _mm_add_ps(fix0,tx);
 535             fiy0             = _mm_add_ps(fiy0,ty);
 536             fiz0             = _mm_add_ps(fiz0,tz);
 537
 538             fjx0             = _mm_add_ps(fjx0,tx);
 539             fjy0             = _mm_add_ps(fjy0,ty);
 540             fjz0             = _mm_add_ps(fjz0,tz);
 541
 542             /**************************
 543              * CALCULATE INTERACTIONS *
 544              **************************/
 545
 546             r10              = _mm_mul_ps(rsq10,rinv10);
 547             r10              = _mm_andnot_ps(dummy_mask,r10);
 548
 549             /* Compute parameters for interactions between i and j atoms */
 550             qq10             = _mm_mul_ps(iq1,jq0);
 551
 552             /* EWALD ELECTROSTATICS */
 553
 554             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 555             ewrt             = _mm_mul_ps(r10,ewtabscale);
 556             ewitab           = _mm_cvttps_epi32(ewrt);
 557             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 558             ewitab           = _mm_slli_epi32(ewitab,2);
 559             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 560             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 561             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 562             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 563             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 564             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 565             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 566             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 567             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 568
 569             /* Update potential sum for this i atom from the interaction with this j atom. */
 570             velec            = _mm_andnot_ps(dummy_mask,velec);
 571             velecsum         = _mm_add_ps(velecsum,velec);
 572
 573             fscal            = felec;
 574
 575             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 576
 577             /* Calculate temporary vectorial force */
 578             tx               = _mm_mul_ps(fscal,dx10);
 579             ty               = _mm_mul_ps(fscal,dy10);
 580             tz               = _mm_mul_ps(fscal,dz10);
 581
 582             /* Update vectorial force */
 583             fix1             = _mm_add_ps(fix1,tx);
 584             fiy1             = _mm_add_ps(fiy1,ty);
 585             fiz1             = _mm_add_ps(fiz1,tz);
 586
 587             fjx0             = _mm_add_ps(fjx0,tx);
 588             fjy0             = _mm_add_ps(fjy0,ty);
 589             fjz0             = _mm_add_ps(fjz0,tz);
 590
 591             /**************************
 592              * CALCULATE INTERACTIONS *
 593              **************************/
 594
 595             r20              = _mm_mul_ps(rsq20,rinv20);
 596             r20              = _mm_andnot_ps(dummy_mask,r20);
 597
 598             /* Compute parameters for interactions between i and j atoms */
 599             qq20             = _mm_mul_ps(iq2,jq0);
 600
 601             /* EWALD ELECTROSTATICS */
 602
 603             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 604             ewrt             = _mm_mul_ps(r20,ewtabscale);
 605             ewitab           = _mm_cvttps_epi32(ewrt);
 606             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 607             ewitab           = _mm_slli_epi32(ewitab,2);
 608             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 609             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 610             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 611             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 612             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 613             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 614             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 615             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 616             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 617
 618             /* Update potential sum for this i atom from the interaction with this j atom. */
 619             velec            = _mm_andnot_ps(dummy_mask,velec);
 620             velecsum         = _mm_add_ps(velecsum,velec);
 621
 622             fscal            = felec;
 623
 624             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 625
 626             /* Calculate temporary vectorial force */
 627             tx               = _mm_mul_ps(fscal,dx20);
 628             ty               = _mm_mul_ps(fscal,dy20);
 629             tz               = _mm_mul_ps(fscal,dz20);
 630
 631             /* Update vectorial force */
 632             fix2             = _mm_add_ps(fix2,tx);
 633             fiy2             = _mm_add_ps(fiy2,ty);
 634             fiz2             = _mm_add_ps(fiz2,tz);
 635
 636             fjx0             = _mm_add_ps(fjx0,tx);
 637             fjy0             = _mm_add_ps(fjy0,ty);
 638             fjz0             = _mm_add_ps(fjz0,tz);
 639
 640             /**************************
 641              * CALCULATE INTERACTIONS *
 642              **************************/
 643
 644             r30              = _mm_mul_ps(rsq30,rinv30);
 645             r30              = _mm_andnot_ps(dummy_mask,r30);
 646
 647             /* Compute parameters for interactions between i and j atoms */
 648             qq30             = _mm_mul_ps(iq3,jq0);
 649
 650             /* EWALD ELECTROSTATICS */
 651
 652             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 653             ewrt             = _mm_mul_ps(r30,ewtabscale);
 654             ewitab           = _mm_cvttps_epi32(ewrt);
 655             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 656             ewitab           = _mm_slli_epi32(ewitab,2);
 657             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 658             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 659             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 660             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 661             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 662             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 663             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 664             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 665             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 666
 667             /* Update potential sum for this i atom from the interaction with this j atom. */
 668             velec            = _mm_andnot_ps(dummy_mask,velec);
 669             velecsum         = _mm_add_ps(velecsum,velec);
 670
 671             fscal            = felec;
 672
 673             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 674
 675             /* Calculate temporary vectorial force */
 676             tx               = _mm_mul_ps(fscal,dx30);
 677             ty               = _mm_mul_ps(fscal,dy30);
 678             tz               = _mm_mul_ps(fscal,dz30);
 679
 680             /* Update vectorial force */
 681             fix3             = _mm_add_ps(fix3,tx);
 682             fiy3             = _mm_add_ps(fiy3,ty);
 683             fiz3             = _mm_add_ps(fiz3,tz);
 684
 685             fjx0             = _mm_add_ps(fjx0,tx);
 686             fjy0             = _mm_add_ps(fjy0,ty);
 687             fjz0             = _mm_add_ps(fjz0,tz);
 688
 689             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 690             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 691             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 692             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 693
 694             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 695
 696             /* Inner loop uses 158 flops */
 697         }
 698
 699         /* End of innermost loop */
 700
 701         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 702                                               f+i_coord_offset,fshift+i_shift_offset);
 703
 704         ggid                        = gid[iidx];
 705         /* Update potential energies */
 706         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 707         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 708
 709         /* Increment number of inner iterations */
 710         inneriter                  += j_index_end - j_index_start;
 711
 712         /* Outer loop uses 26 flops */
 713     }
 714
 715     /* Increment number of outer iterations */
 716     outeriter        += nri;
 717
 718     /* Update outer/inner flops */
 719
 720     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*158);
 721 }
 722 /*
 723  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_sse2_single
 724  * Electrostatics interaction: Ewald
 725  * VdW interaction:            LennardJones
 726  * Geometry:                   Water4-Particle
 727  * Calculate force/pot:        Force
 728  */
 729 void
 730 nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_sse2_single
 731                     (t_nblist                    * gmx_restrict       nlist,
 732                      rvec                        * gmx_restrict          xx,
 733                      rvec                        * gmx_restrict          ff,
 734                      struct t_forcerec           * gmx_restrict          fr,
 735                      t_mdatoms                   * gmx_restrict     mdatoms,
 736                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 737                      t_nrnb                      * gmx_restrict        nrnb)
 738 {
 739     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 740      * just 0 for non-waters.
 741      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 742      * jnr indices corresponding to data put in the four positions in the SIMD register.
 743      */
 744     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 745     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 746     int              jnrA,jnrB,jnrC,jnrD;
 747     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 748     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 749     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 750     real             rcutoff_scalar;
 751     real             *shiftvec,*fshift,*x,*f;
 752     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 753     real             scratch[4*DIM];
 754     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 755     int              vdwioffset0;
 756     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 757     int              vdwioffset1;
 758     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 759     int              vdwioffset2;
 760     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 761     int              vdwioffset3;
 762     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 763     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 764     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 765     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 766     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 767     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 768     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 769     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 770     real             *charge;
 771     int              nvdwtype;
 772     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 773     int              *vdwtype;
 774     real             *vdwparam;
 775     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 776     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 777     __m128i          ewitab;
 778     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 779     real             *ewtab;
 780     __m128           dummy_mask,cutoff_mask;
 781     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 782     __m128           one     = _mm_set1_ps(1.0);
 783     __m128           two     = _mm_set1_ps(2.0);
 784     x                = xx[0];
 785     f                = ff[0];
 786
 787     nri              = nlist->nri;
 788     iinr             = nlist->iinr;
 789     jindex           = nlist->jindex;
 790     jjnr             = nlist->jjnr;
 791     shiftidx         = nlist->shift;
 792     gid              = nlist->gid;
 793     shiftvec         = fr->shift_vec[0];
 794     fshift           = fr->fshift[0];
 795     facel            = _mm_set1_ps(fr->ic->epsfac);
 796     charge           = mdatoms->chargeA;
 797     nvdwtype         = fr->ntype;
 798     vdwparam         = fr->nbfp;
 799     vdwtype          = mdatoms->typeA;
 800
 801     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 802     ewtab            = fr->ic->tabq_coul_F;
 803     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 804     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 805
 806     /* Setup water-specific parameters */
 807     inr              = nlist->iinr[0];
 808     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 809     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 810     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 811     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 812
 813     /* Avoid stupid compiler warnings */
 814     jnrA = jnrB = jnrC = jnrD = 0;
 815     j_coord_offsetA = 0;
 816     j_coord_offsetB = 0;
 817     j_coord_offsetC = 0;
 818     j_coord_offsetD = 0;
 819
 820     outeriter        = 0;
 821     inneriter        = 0;
 822
 823     for(iidx=0;iidx<4*DIM;iidx++)
 824     {
 825         scratch[iidx] = 0.0;
 826     }
 827
 828     /* Start outer loop over neighborlists */
 829     for(iidx=0; iidx<nri; iidx++)
 830     {
 831         /* Load shift vector for this list */
 832         i_shift_offset   = DIM*shiftidx[iidx];
 833
 834         /* Load limits for loop over neighbors */
 835         j_index_start    = jindex[iidx];
 836         j_index_end      = jindex[iidx+1];
 837
 838         /* Get outer coordinate index */
 839         inr              = iinr[iidx];
 840         i_coord_offset   = DIM*inr;
 841
 842         /* Load i particle coords and add shift vector */
 843         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 844                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 845
 846         fix0             = _mm_setzero_ps();
 847         fiy0             = _mm_setzero_ps();
 848         fiz0             = _mm_setzero_ps();
 849         fix1             = _mm_setzero_ps();
 850         fiy1             = _mm_setzero_ps();
 851         fiz1             = _mm_setzero_ps();
 852         fix2             = _mm_setzero_ps();
 853         fiy2             = _mm_setzero_ps();
 854         fiz2             = _mm_setzero_ps();
 855         fix3             = _mm_setzero_ps();
 856         fiy3             = _mm_setzero_ps();
 857         fiz3             = _mm_setzero_ps();
 858
 859         /* Start inner kernel loop */
 860         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 861         {
 862
 863             /* Get j neighbor index, and coordinate index */
 864             jnrA             = jjnr[jidx];
 865             jnrB             = jjnr[jidx+1];
 866             jnrC             = jjnr[jidx+2];
 867             jnrD             = jjnr[jidx+3];
 868             j_coord_offsetA  = DIM*jnrA;
 869             j_coord_offsetB  = DIM*jnrB;
 870             j_coord_offsetC  = DIM*jnrC;
 871             j_coord_offsetD  = DIM*jnrD;
 872
 873             /* load j atom coordinates */
 874             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 875                                               x+j_coord_offsetC,x+j_coord_offsetD,
 876                                               &jx0,&jy0,&jz0);
 877
 878             /* Calculate displacement vector */
 879             dx00             = _mm_sub_ps(ix0,jx0);
 880             dy00             = _mm_sub_ps(iy0,jy0);
 881             dz00             = _mm_sub_ps(iz0,jz0);
 882             dx10             = _mm_sub_ps(ix1,jx0);
 883             dy10             = _mm_sub_ps(iy1,jy0);
 884             dz10             = _mm_sub_ps(iz1,jz0);
 885             dx20             = _mm_sub_ps(ix2,jx0);
 886             dy20             = _mm_sub_ps(iy2,jy0);
 887             dz20             = _mm_sub_ps(iz2,jz0);
 888             dx30             = _mm_sub_ps(ix3,jx0);
 889             dy30             = _mm_sub_ps(iy3,jy0);
 890             dz30             = _mm_sub_ps(iz3,jz0);
 891
 892             /* Calculate squared distance and things based on it */
 893             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 894             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 895             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 896             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 897
 898             rinv10           = sse2_invsqrt_f(rsq10);
 899             rinv20           = sse2_invsqrt_f(rsq20);
 900             rinv30           = sse2_invsqrt_f(rsq30);
 901
 902             rinvsq00         = sse2_inv_f(rsq00);
 903             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 904             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 905             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 906
 907             /* Load parameters for j particles */
 908             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 909                                                               charge+jnrC+0,charge+jnrD+0);
 910             vdwjidx0A        = 2*vdwtype[jnrA+0];
 911             vdwjidx0B        = 2*vdwtype[jnrB+0];
 912             vdwjidx0C        = 2*vdwtype[jnrC+0];
 913             vdwjidx0D        = 2*vdwtype[jnrD+0];
 914
 915             fjx0             = _mm_setzero_ps();
 916             fjy0             = _mm_setzero_ps();
 917             fjz0             = _mm_setzero_ps();
 918
 919             /**************************
 920              * CALCULATE INTERACTIONS *
 921              **************************/
 922
 923             /* Compute parameters for interactions between i and j atoms */
 924             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 925                                          vdwparam+vdwioffset0+vdwjidx0B,
 926                                          vdwparam+vdwioffset0+vdwjidx0C,
 927                                          vdwparam+vdwioffset0+vdwjidx0D,
 928                                          &c6_00,&c12_00);
 929
 930             /* LENNARD-JONES DISPERSION/REPULSION */
 931
 932             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 933             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 934
 935             fscal            = fvdw;
 936
 937             /* Calculate temporary vectorial force */
 938             tx               = _mm_mul_ps(fscal,dx00);
 939             ty               = _mm_mul_ps(fscal,dy00);
 940             tz               = _mm_mul_ps(fscal,dz00);
 941
 942             /* Update vectorial force */
 943             fix0             = _mm_add_ps(fix0,tx);
 944             fiy0             = _mm_add_ps(fiy0,ty);
 945             fiz0             = _mm_add_ps(fiz0,tz);
 946
 947             fjx0             = _mm_add_ps(fjx0,tx);
 948             fjy0             = _mm_add_ps(fjy0,ty);
 949             fjz0             = _mm_add_ps(fjz0,tz);
 950
 951             /**************************
 952              * CALCULATE INTERACTIONS *
 953              **************************/
 954
 955             r10              = _mm_mul_ps(rsq10,rinv10);
 956
 957             /* Compute parameters for interactions between i and j atoms */
 958             qq10             = _mm_mul_ps(iq1,jq0);
 959
 960             /* EWALD ELECTROSTATICS */
 961
 962             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 963             ewrt             = _mm_mul_ps(r10,ewtabscale);
 964             ewitab           = _mm_cvttps_epi32(ewrt);
 965             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 966             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 967                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 968                                          &ewtabF,&ewtabFn);
 969             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 970             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 971
 972             fscal            = felec;
 973
 974             /* Calculate temporary vectorial force */
 975             tx               = _mm_mul_ps(fscal,dx10);
 976             ty               = _mm_mul_ps(fscal,dy10);
 977             tz               = _mm_mul_ps(fscal,dz10);
 978
 979             /* Update vectorial force */
 980             fix1             = _mm_add_ps(fix1,tx);
 981             fiy1             = _mm_add_ps(fiy1,ty);
 982             fiz1             = _mm_add_ps(fiz1,tz);
 983
 984             fjx0             = _mm_add_ps(fjx0,tx);
 985             fjy0             = _mm_add_ps(fjy0,ty);
 986             fjz0             = _mm_add_ps(fjz0,tz);
 987
 988             /**************************
 989              * CALCULATE INTERACTIONS *
 990              **************************/
 991
 992             r20              = _mm_mul_ps(rsq20,rinv20);
 993
 994             /* Compute parameters for interactions between i and j atoms */
 995             qq20             = _mm_mul_ps(iq2,jq0);
 996
 997             /* EWALD ELECTROSTATICS */
 998
 999             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1000             ewrt             = _mm_mul_ps(r20,ewtabscale);
1001             ewitab           = _mm_cvttps_epi32(ewrt);
1002             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1003             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1004                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1005                                          &ewtabF,&ewtabFn);
1006             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1007             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1008
1009             fscal            = felec;
1010
1011             /* Calculate temporary vectorial force */
1012             tx               = _mm_mul_ps(fscal,dx20);
1013             ty               = _mm_mul_ps(fscal,dy20);
1014             tz               = _mm_mul_ps(fscal,dz20);
1015
1016             /* Update vectorial force */
1017             fix2             = _mm_add_ps(fix2,tx);
1018             fiy2             = _mm_add_ps(fiy2,ty);
1019             fiz2             = _mm_add_ps(fiz2,tz);
1020
1021             fjx0             = _mm_add_ps(fjx0,tx);
1022             fjy0             = _mm_add_ps(fjy0,ty);
1023             fjz0             = _mm_add_ps(fjz0,tz);
1024
1025             /**************************
1026              * CALCULATE INTERACTIONS *
1027              **************************/
1028
1029             r30              = _mm_mul_ps(rsq30,rinv30);
1030
1031             /* Compute parameters for interactions between i and j atoms */
1032             qq30             = _mm_mul_ps(iq3,jq0);
1033
1034             /* EWALD ELECTROSTATICS */
1035
1036             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1037             ewrt             = _mm_mul_ps(r30,ewtabscale);
1038             ewitab           = _mm_cvttps_epi32(ewrt);
1039             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1040             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1041                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1042                                          &ewtabF,&ewtabFn);
1043             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1044             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1045
1046             fscal            = felec;
1047
1048             /* Calculate temporary vectorial force */
1049             tx               = _mm_mul_ps(fscal,dx30);
1050             ty               = _mm_mul_ps(fscal,dy30);
1051             tz               = _mm_mul_ps(fscal,dz30);
1052
1053             /* Update vectorial force */
1054             fix3             = _mm_add_ps(fix3,tx);
1055             fiy3             = _mm_add_ps(fiy3,ty);
1056             fiz3             = _mm_add_ps(fiz3,tz);
1057
1058             fjx0             = _mm_add_ps(fjx0,tx);
1059             fjy0             = _mm_add_ps(fjy0,ty);
1060             fjz0             = _mm_add_ps(fjz0,tz);
1061
1062             fjptrA             = f+j_coord_offsetA;
1063             fjptrB             = f+j_coord_offsetB;
1064             fjptrC             = f+j_coord_offsetC;
1065             fjptrD             = f+j_coord_offsetD;
1066
1067             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1068
1069             /* Inner loop uses 135 flops */
1070         }
1071
1072         if(jidx<j_index_end)
1073         {
1074
1075             /* Get j neighbor index, and coordinate index */
1076             jnrlistA         = jjnr[jidx];
1077             jnrlistB         = jjnr[jidx+1];
1078             jnrlistC         = jjnr[jidx+2];
1079             jnrlistD         = jjnr[jidx+3];
1080             /* Sign of each element will be negative for non-real atoms.
1081              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1082              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1083              */
1084             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1085             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1086             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1087             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1088             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1089             j_coord_offsetA  = DIM*jnrA;
1090             j_coord_offsetB  = DIM*jnrB;
1091             j_coord_offsetC  = DIM*jnrC;
1092             j_coord_offsetD  = DIM*jnrD;
1093
1094             /* load j atom coordinates */
1095             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1096                                               x+j_coord_offsetC,x+j_coord_offsetD,
1097                                               &jx0,&jy0,&jz0);
1098
1099             /* Calculate displacement vector */
1100             dx00             = _mm_sub_ps(ix0,jx0);
1101             dy00             = _mm_sub_ps(iy0,jy0);
1102             dz00             = _mm_sub_ps(iz0,jz0);
1103             dx10             = _mm_sub_ps(ix1,jx0);
1104             dy10             = _mm_sub_ps(iy1,jy0);
1105             dz10             = _mm_sub_ps(iz1,jz0);
1106             dx20             = _mm_sub_ps(ix2,jx0);
1107             dy20             = _mm_sub_ps(iy2,jy0);
1108             dz20             = _mm_sub_ps(iz2,jz0);
1109             dx30             = _mm_sub_ps(ix3,jx0);
1110             dy30             = _mm_sub_ps(iy3,jy0);
1111             dz30             = _mm_sub_ps(iz3,jz0);
1112
1113             /* Calculate squared distance and things based on it */
1114             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1115             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1116             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1117             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1118
1119             rinv10           = sse2_invsqrt_f(rsq10);
1120             rinv20           = sse2_invsqrt_f(rsq20);
1121             rinv30           = sse2_invsqrt_f(rsq30);
1122
1123             rinvsq00         = sse2_inv_f(rsq00);
1124             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1125             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1126             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1127
1128             /* Load parameters for j particles */
1129             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1130                                                               charge+jnrC+0,charge+jnrD+0);
1131             vdwjidx0A        = 2*vdwtype[jnrA+0];
1132             vdwjidx0B        = 2*vdwtype[jnrB+0];
1133             vdwjidx0C        = 2*vdwtype[jnrC+0];
1134             vdwjidx0D        = 2*vdwtype[jnrD+0];
1135
1136             fjx0             = _mm_setzero_ps();
1137             fjy0             = _mm_setzero_ps();
1138             fjz0             = _mm_setzero_ps();
1139
1140             /**************************
1141              * CALCULATE INTERACTIONS *
1142              **************************/
1143
1144             /* Compute parameters for interactions between i and j atoms */
1145             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1146                                          vdwparam+vdwioffset0+vdwjidx0B,
1147                                          vdwparam+vdwioffset0+vdwjidx0C,
1148                                          vdwparam+vdwioffset0+vdwjidx0D,
1149                                          &c6_00,&c12_00);
1150
1151             /* LENNARD-JONES DISPERSION/REPULSION */
1152
1153             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1154             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1155
1156             fscal            = fvdw;
1157
1158             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1159
1160             /* Calculate temporary vectorial force */
1161             tx               = _mm_mul_ps(fscal,dx00);
1162             ty               = _mm_mul_ps(fscal,dy00);
1163             tz               = _mm_mul_ps(fscal,dz00);
1164
1165             /* Update vectorial force */
1166             fix0             = _mm_add_ps(fix0,tx);
1167             fiy0             = _mm_add_ps(fiy0,ty);
1168             fiz0             = _mm_add_ps(fiz0,tz);
1169
1170             fjx0             = _mm_add_ps(fjx0,tx);
1171             fjy0             = _mm_add_ps(fjy0,ty);
1172             fjz0             = _mm_add_ps(fjz0,tz);
1173
1174             /**************************
1175              * CALCULATE INTERACTIONS *
1176              **************************/
1177
1178             r10              = _mm_mul_ps(rsq10,rinv10);
1179             r10              = _mm_andnot_ps(dummy_mask,r10);
1180
1181             /* Compute parameters for interactions between i and j atoms */
1182             qq10             = _mm_mul_ps(iq1,jq0);
1183
1184             /* EWALD ELECTROSTATICS */
1185
1186             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1187             ewrt             = _mm_mul_ps(r10,ewtabscale);
1188             ewitab           = _mm_cvttps_epi32(ewrt);
1189             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1190             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1191                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1192                                          &ewtabF,&ewtabFn);
1193             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1194             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1195
1196             fscal            = felec;
1197
1198             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1199
1200             /* Calculate temporary vectorial force */
1201             tx               = _mm_mul_ps(fscal,dx10);
1202             ty               = _mm_mul_ps(fscal,dy10);
1203             tz               = _mm_mul_ps(fscal,dz10);
1204
1205             /* Update vectorial force */
1206             fix1             = _mm_add_ps(fix1,tx);
1207             fiy1             = _mm_add_ps(fiy1,ty);
1208             fiz1             = _mm_add_ps(fiz1,tz);
1209
1210             fjx0             = _mm_add_ps(fjx0,tx);
1211             fjy0             = _mm_add_ps(fjy0,ty);
1212             fjz0             = _mm_add_ps(fjz0,tz);
1213
1214             /**************************
1215              * CALCULATE INTERACTIONS *
1216              **************************/
1217
1218             r20              = _mm_mul_ps(rsq20,rinv20);
1219             r20              = _mm_andnot_ps(dummy_mask,r20);
1220
1221             /* Compute parameters for interactions between i and j atoms */
1222             qq20             = _mm_mul_ps(iq2,jq0);
1223
1224             /* EWALD ELECTROSTATICS */
1225
1226             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1227             ewrt             = _mm_mul_ps(r20,ewtabscale);
1228             ewitab           = _mm_cvttps_epi32(ewrt);
1229             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1230             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1231                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1232                                          &ewtabF,&ewtabFn);
1233             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1234             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1235
1236             fscal            = felec;
1237
1238             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1239
1240             /* Calculate temporary vectorial force */
1241             tx               = _mm_mul_ps(fscal,dx20);
1242             ty               = _mm_mul_ps(fscal,dy20);
1243             tz               = _mm_mul_ps(fscal,dz20);
1244
1245             /* Update vectorial force */
1246             fix2             = _mm_add_ps(fix2,tx);
1247             fiy2             = _mm_add_ps(fiy2,ty);
1248             fiz2             = _mm_add_ps(fiz2,tz);
1249
1250             fjx0             = _mm_add_ps(fjx0,tx);
1251             fjy0             = _mm_add_ps(fjy0,ty);
1252             fjz0             = _mm_add_ps(fjz0,tz);
1253
1254             /**************************
1255              * CALCULATE INTERACTIONS *
1256              **************************/
1257
1258             r30              = _mm_mul_ps(rsq30,rinv30);
1259             r30              = _mm_andnot_ps(dummy_mask,r30);
1260
1261             /* Compute parameters for interactions between i and j atoms */
1262             qq30             = _mm_mul_ps(iq3,jq0);
1263
1264             /* EWALD ELECTROSTATICS */
1265
1266             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1267             ewrt             = _mm_mul_ps(r30,ewtabscale);
1268             ewitab           = _mm_cvttps_epi32(ewrt);
1269             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1270             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1271                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1272                                          &ewtabF,&ewtabFn);
1273             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1274             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1275
1276             fscal            = felec;
1277
1278             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1279
1280             /* Calculate temporary vectorial force */
1281             tx               = _mm_mul_ps(fscal,dx30);
1282             ty               = _mm_mul_ps(fscal,dy30);
1283             tz               = _mm_mul_ps(fscal,dz30);
1284
1285             /* Update vectorial force */
1286             fix3             = _mm_add_ps(fix3,tx);
1287             fiy3             = _mm_add_ps(fiy3,ty);
1288             fiz3             = _mm_add_ps(fiz3,tz);
1289
1290             fjx0             = _mm_add_ps(fjx0,tx);
1291             fjy0             = _mm_add_ps(fjy0,ty);
1292             fjz0             = _mm_add_ps(fjz0,tz);
1293
1294             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1295             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1296             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1297             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1298
1299             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1300
1301             /* Inner loop uses 138 flops */
1302         }
1303
1304         /* End of innermost loop */
1305
1306         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1307                                               f+i_coord_offset,fshift+i_shift_offset);
1308
1309         /* Increment number of inner iterations */
1310         inneriter                  += j_index_end - j_index_start;
1311
1312         /* Outer loop uses 24 flops */
1313     }
1314
1315     /* Increment number of outer iterations */
1316     outeriter        += nri;
1317
1318     /* Update outer/inner flops */
1319
1320     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*138);
1321 }