src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEw_VdwLJ_GeomW4P1_sse4_1_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, 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 sse4_1_single kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gmx_math_x86_sse4_1_single.h"
  50 #include "kernelutil_x86_sse4_1_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW4P1_VF_sse4_1_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwLJ_GeomW4P1_VF_sse4_1_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwioffset3;
  92     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  93     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  94     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  95     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  96     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  97     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  98     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  99     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     int              nvdwtype;
 102     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 106     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 107     __m128i          ewitab;
 108     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 109     real             *ewtab;
 110     __m128           dummy_mask,cutoff_mask;
 111     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 112     __m128           one     = _mm_set1_ps(1.0);
 113     __m128           two     = _mm_set1_ps(2.0);
 114     x                = xx[0];
 115     f                = ff[0];
 116
 117     nri              = nlist->nri;
 118     iinr             = nlist->iinr;
 119     jindex           = nlist->jindex;
 120     jjnr             = nlist->jjnr;
 121     shiftidx         = nlist->shift;
 122     gid              = nlist->gid;
 123     shiftvec         = fr->shift_vec[0];
 124     fshift           = fr->fshift[0];
 125     facel            = _mm_set1_ps(fr->epsfac);
 126     charge           = mdatoms->chargeA;
 127     nvdwtype         = fr->ntype;
 128     vdwparam         = fr->nbfp;
 129     vdwtype          = mdatoms->typeA;
 130
 131     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 132     ewtab            = fr->ic->tabq_coul_FDV0;
 133     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 134     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 135
 136     /* Setup water-specific parameters */
 137     inr              = nlist->iinr[0];
 138     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 139     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 140     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 141     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 142
 143     /* Avoid stupid compiler warnings */
 144     jnrA = jnrB = jnrC = jnrD = 0;
 145     j_coord_offsetA = 0;
 146     j_coord_offsetB = 0;
 147     j_coord_offsetC = 0;
 148     j_coord_offsetD = 0;
 149
 150     outeriter        = 0;
 151     inneriter        = 0;
 152
 153     for(iidx=0;iidx<4*DIM;iidx++)
 154     {
 155         scratch[iidx] = 0.0;
 156     }
 157
 158     /* Start outer loop over neighborlists */
 159     for(iidx=0; iidx<nri; iidx++)
 160     {
 161         /* Load shift vector for this list */
 162         i_shift_offset   = DIM*shiftidx[iidx];
 163
 164         /* Load limits for loop over neighbors */
 165         j_index_start    = jindex[iidx];
 166         j_index_end      = jindex[iidx+1];
 167
 168         /* Get outer coordinate index */
 169         inr              = iinr[iidx];
 170         i_coord_offset   = DIM*inr;
 171
 172         /* Load i particle coords and add shift vector */
 173         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 174                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 175
 176         fix0             = _mm_setzero_ps();
 177         fiy0             = _mm_setzero_ps();
 178         fiz0             = _mm_setzero_ps();
 179         fix1             = _mm_setzero_ps();
 180         fiy1             = _mm_setzero_ps();
 181         fiz1             = _mm_setzero_ps();
 182         fix2             = _mm_setzero_ps();
 183         fiy2             = _mm_setzero_ps();
 184         fiz2             = _mm_setzero_ps();
 185         fix3             = _mm_setzero_ps();
 186         fiy3             = _mm_setzero_ps();
 187         fiz3             = _mm_setzero_ps();
 188
 189         /* Reset potential sums */
 190         velecsum         = _mm_setzero_ps();
 191         vvdwsum          = _mm_setzero_ps();
 192
 193         /* Start inner kernel loop */
 194         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 195         {
 196
 197             /* Get j neighbor index, and coordinate index */
 198             jnrA             = jjnr[jidx];
 199             jnrB             = jjnr[jidx+1];
 200             jnrC             = jjnr[jidx+2];
 201             jnrD             = jjnr[jidx+3];
 202             j_coord_offsetA  = DIM*jnrA;
 203             j_coord_offsetB  = DIM*jnrB;
 204             j_coord_offsetC  = DIM*jnrC;
 205             j_coord_offsetD  = DIM*jnrD;
 206
 207             /* load j atom coordinates */
 208             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 209                                               x+j_coord_offsetC,x+j_coord_offsetD,
 210                                               &jx0,&jy0,&jz0);
 211
 212             /* Calculate displacement vector */
 213             dx00             = _mm_sub_ps(ix0,jx0);
 214             dy00             = _mm_sub_ps(iy0,jy0);
 215             dz00             = _mm_sub_ps(iz0,jz0);
 216             dx10             = _mm_sub_ps(ix1,jx0);
 217             dy10             = _mm_sub_ps(iy1,jy0);
 218             dz10             = _mm_sub_ps(iz1,jz0);
 219             dx20             = _mm_sub_ps(ix2,jx0);
 220             dy20             = _mm_sub_ps(iy2,jy0);
 221             dz20             = _mm_sub_ps(iz2,jz0);
 222             dx30             = _mm_sub_ps(ix3,jx0);
 223             dy30             = _mm_sub_ps(iy3,jy0);
 224             dz30             = _mm_sub_ps(iz3,jz0);
 225
 226             /* Calculate squared distance and things based on it */
 227             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 228             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 229             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 230             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 231
 232             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 233             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 234             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 235
 236             rinvsq00         = gmx_mm_inv_ps(rsq00);
 237             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 238             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 239             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 240
 241             /* Load parameters for j particles */
 242             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 243                                                               charge+jnrC+0,charge+jnrD+0);
 244             vdwjidx0A        = 2*vdwtype[jnrA+0];
 245             vdwjidx0B        = 2*vdwtype[jnrB+0];
 246             vdwjidx0C        = 2*vdwtype[jnrC+0];
 247             vdwjidx0D        = 2*vdwtype[jnrD+0];
 248
 249             fjx0             = _mm_setzero_ps();
 250             fjy0             = _mm_setzero_ps();
 251             fjz0             = _mm_setzero_ps();
 252
 253             /**************************
 254              * CALCULATE INTERACTIONS *
 255              **************************/
 256
 257             /* Compute parameters for interactions between i and j atoms */
 258             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 259                                          vdwparam+vdwioffset0+vdwjidx0B,
 260                                          vdwparam+vdwioffset0+vdwjidx0C,
 261                                          vdwparam+vdwioffset0+vdwjidx0D,
 262                                          &c6_00,&c12_00);
 263
 264             /* LENNARD-JONES DISPERSION/REPULSION */
 265
 266             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 267             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 268             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 269             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 270             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 271
 272             /* Update potential sum for this i atom from the interaction with this j atom. */
 273             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 274
 275             fscal            = fvdw;
 276
 277             /* Calculate temporary vectorial force */
 278             tx               = _mm_mul_ps(fscal,dx00);
 279             ty               = _mm_mul_ps(fscal,dy00);
 280             tz               = _mm_mul_ps(fscal,dz00);
 281
 282             /* Update vectorial force */
 283             fix0             = _mm_add_ps(fix0,tx);
 284             fiy0             = _mm_add_ps(fiy0,ty);
 285             fiz0             = _mm_add_ps(fiz0,tz);
 286
 287             fjx0             = _mm_add_ps(fjx0,tx);
 288             fjy0             = _mm_add_ps(fjy0,ty);
 289             fjz0             = _mm_add_ps(fjz0,tz);
 290
 291             /**************************
 292              * CALCULATE INTERACTIONS *
 293              **************************/
 294
 295             r10              = _mm_mul_ps(rsq10,rinv10);
 296
 297             /* Compute parameters for interactions between i and j atoms */
 298             qq10             = _mm_mul_ps(iq1,jq0);
 299
 300             /* EWALD ELECTROSTATICS */
 301
 302             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 303             ewrt             = _mm_mul_ps(r10,ewtabscale);
 304             ewitab           = _mm_cvttps_epi32(ewrt);
 305             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 306             ewitab           = _mm_slli_epi32(ewitab,2);
 307             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 308             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 309             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 310             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 311             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 312             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 313             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 314             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 315             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 316
 317             /* Update potential sum for this i atom from the interaction with this j atom. */
 318             velecsum         = _mm_add_ps(velecsum,velec);
 319
 320             fscal            = felec;
 321
 322             /* Calculate temporary vectorial force */
 323             tx               = _mm_mul_ps(fscal,dx10);
 324             ty               = _mm_mul_ps(fscal,dy10);
 325             tz               = _mm_mul_ps(fscal,dz10);
 326
 327             /* Update vectorial force */
 328             fix1             = _mm_add_ps(fix1,tx);
 329             fiy1             = _mm_add_ps(fiy1,ty);
 330             fiz1             = _mm_add_ps(fiz1,tz);
 331
 332             fjx0             = _mm_add_ps(fjx0,tx);
 333             fjy0             = _mm_add_ps(fjy0,ty);
 334             fjz0             = _mm_add_ps(fjz0,tz);
 335
 336             /**************************
 337              * CALCULATE INTERACTIONS *
 338              **************************/
 339
 340             r20              = _mm_mul_ps(rsq20,rinv20);
 341
 342             /* Compute parameters for interactions between i and j atoms */
 343             qq20             = _mm_mul_ps(iq2,jq0);
 344
 345             /* EWALD ELECTROSTATICS */
 346
 347             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 348             ewrt             = _mm_mul_ps(r20,ewtabscale);
 349             ewitab           = _mm_cvttps_epi32(ewrt);
 350             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 351             ewitab           = _mm_slli_epi32(ewitab,2);
 352             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 353             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 354             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 355             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 356             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 357             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 358             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 359             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 360             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 361
 362             /* Update potential sum for this i atom from the interaction with this j atom. */
 363             velecsum         = _mm_add_ps(velecsum,velec);
 364
 365             fscal            = felec;
 366
 367             /* Calculate temporary vectorial force */
 368             tx               = _mm_mul_ps(fscal,dx20);
 369             ty               = _mm_mul_ps(fscal,dy20);
 370             tz               = _mm_mul_ps(fscal,dz20);
 371
 372             /* Update vectorial force */
 373             fix2             = _mm_add_ps(fix2,tx);
 374             fiy2             = _mm_add_ps(fiy2,ty);
 375             fiz2             = _mm_add_ps(fiz2,tz);
 376
 377             fjx0             = _mm_add_ps(fjx0,tx);
 378             fjy0             = _mm_add_ps(fjy0,ty);
 379             fjz0             = _mm_add_ps(fjz0,tz);
 380
 381             /**************************
 382              * CALCULATE INTERACTIONS *
 383              **************************/
 384
 385             r30              = _mm_mul_ps(rsq30,rinv30);
 386
 387             /* Compute parameters for interactions between i and j atoms */
 388             qq30             = _mm_mul_ps(iq3,jq0);
 389
 390             /* EWALD ELECTROSTATICS */
 391
 392             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 393             ewrt             = _mm_mul_ps(r30,ewtabscale);
 394             ewitab           = _mm_cvttps_epi32(ewrt);
 395             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 396             ewitab           = _mm_slli_epi32(ewitab,2);
 397             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 398             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 399             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 400             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 401             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 402             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 403             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 404             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 405             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 406
 407             /* Update potential sum for this i atom from the interaction with this j atom. */
 408             velecsum         = _mm_add_ps(velecsum,velec);
 409
 410             fscal            = felec;
 411
 412             /* Calculate temporary vectorial force */
 413             tx               = _mm_mul_ps(fscal,dx30);
 414             ty               = _mm_mul_ps(fscal,dy30);
 415             tz               = _mm_mul_ps(fscal,dz30);
 416
 417             /* Update vectorial force */
 418             fix3             = _mm_add_ps(fix3,tx);
 419             fiy3             = _mm_add_ps(fiy3,ty);
 420             fiz3             = _mm_add_ps(fiz3,tz);
 421
 422             fjx0             = _mm_add_ps(fjx0,tx);
 423             fjy0             = _mm_add_ps(fjy0,ty);
 424             fjz0             = _mm_add_ps(fjz0,tz);
 425
 426             fjptrA             = f+j_coord_offsetA;
 427             fjptrB             = f+j_coord_offsetB;
 428             fjptrC             = f+j_coord_offsetC;
 429             fjptrD             = f+j_coord_offsetD;
 430
 431             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 432
 433             /* Inner loop uses 155 flops */
 434         }
 435
 436         if(jidx<j_index_end)
 437         {
 438
 439             /* Get j neighbor index, and coordinate index */
 440             jnrlistA         = jjnr[jidx];
 441             jnrlistB         = jjnr[jidx+1];
 442             jnrlistC         = jjnr[jidx+2];
 443             jnrlistD         = jjnr[jidx+3];
 444             /* Sign of each element will be negative for non-real atoms.
 445              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 446              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 447              */
 448             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 449             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 450             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 451             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 452             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 453             j_coord_offsetA  = DIM*jnrA;
 454             j_coord_offsetB  = DIM*jnrB;
 455             j_coord_offsetC  = DIM*jnrC;
 456             j_coord_offsetD  = DIM*jnrD;
 457
 458             /* load j atom coordinates */
 459             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 460                                               x+j_coord_offsetC,x+j_coord_offsetD,
 461                                               &jx0,&jy0,&jz0);
 462
 463             /* Calculate displacement vector */
 464             dx00             = _mm_sub_ps(ix0,jx0);
 465             dy00             = _mm_sub_ps(iy0,jy0);
 466             dz00             = _mm_sub_ps(iz0,jz0);
 467             dx10             = _mm_sub_ps(ix1,jx0);
 468             dy10             = _mm_sub_ps(iy1,jy0);
 469             dz10             = _mm_sub_ps(iz1,jz0);
 470             dx20             = _mm_sub_ps(ix2,jx0);
 471             dy20             = _mm_sub_ps(iy2,jy0);
 472             dz20             = _mm_sub_ps(iz2,jz0);
 473             dx30             = _mm_sub_ps(ix3,jx0);
 474             dy30             = _mm_sub_ps(iy3,jy0);
 475             dz30             = _mm_sub_ps(iz3,jz0);
 476
 477             /* Calculate squared distance and things based on it */
 478             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 479             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 480             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 481             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 482
 483             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 484             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 485             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 486
 487             rinvsq00         = gmx_mm_inv_ps(rsq00);
 488             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 489             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 490             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 491
 492             /* Load parameters for j particles */
 493             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 494                                                               charge+jnrC+0,charge+jnrD+0);
 495             vdwjidx0A        = 2*vdwtype[jnrA+0];
 496             vdwjidx0B        = 2*vdwtype[jnrB+0];
 497             vdwjidx0C        = 2*vdwtype[jnrC+0];
 498             vdwjidx0D        = 2*vdwtype[jnrD+0];
 499
 500             fjx0             = _mm_setzero_ps();
 501             fjy0             = _mm_setzero_ps();
 502             fjz0             = _mm_setzero_ps();
 503
 504             /**************************
 505              * CALCULATE INTERACTIONS *
 506              **************************/
 507
 508             /* Compute parameters for interactions between i and j atoms */
 509             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 510                                          vdwparam+vdwioffset0+vdwjidx0B,
 511                                          vdwparam+vdwioffset0+vdwjidx0C,
 512                                          vdwparam+vdwioffset0+vdwjidx0D,
 513                                          &c6_00,&c12_00);
 514
 515             /* LENNARD-JONES DISPERSION/REPULSION */
 516
 517             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 518             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 519             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 520             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 521             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 522
 523             /* Update potential sum for this i atom from the interaction with this j atom. */
 524             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 525             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 526
 527             fscal            = fvdw;
 528
 529             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 530
 531             /* Calculate temporary vectorial force */
 532             tx               = _mm_mul_ps(fscal,dx00);
 533             ty               = _mm_mul_ps(fscal,dy00);
 534             tz               = _mm_mul_ps(fscal,dz00);
 535
 536             /* Update vectorial force */
 537             fix0             = _mm_add_ps(fix0,tx);
 538             fiy0             = _mm_add_ps(fiy0,ty);
 539             fiz0             = _mm_add_ps(fiz0,tz);
 540
 541             fjx0             = _mm_add_ps(fjx0,tx);
 542             fjy0             = _mm_add_ps(fjy0,ty);
 543             fjz0             = _mm_add_ps(fjz0,tz);
 544
 545             /**************************
 546              * CALCULATE INTERACTIONS *
 547              **************************/
 548
 549             r10              = _mm_mul_ps(rsq10,rinv10);
 550             r10              = _mm_andnot_ps(dummy_mask,r10);
 551
 552             /* Compute parameters for interactions between i and j atoms */
 553             qq10             = _mm_mul_ps(iq1,jq0);
 554
 555             /* EWALD ELECTROSTATICS */
 556
 557             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 558             ewrt             = _mm_mul_ps(r10,ewtabscale);
 559             ewitab           = _mm_cvttps_epi32(ewrt);
 560             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 561             ewitab           = _mm_slli_epi32(ewitab,2);
 562             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 563             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 564             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 565             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 566             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 567             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 568             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 569             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 570             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 571
 572             /* Update potential sum for this i atom from the interaction with this j atom. */
 573             velec            = _mm_andnot_ps(dummy_mask,velec);
 574             velecsum         = _mm_add_ps(velecsum,velec);
 575
 576             fscal            = felec;
 577
 578             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 579
 580             /* Calculate temporary vectorial force */
 581             tx               = _mm_mul_ps(fscal,dx10);
 582             ty               = _mm_mul_ps(fscal,dy10);
 583             tz               = _mm_mul_ps(fscal,dz10);
 584
 585             /* Update vectorial force */
 586             fix1             = _mm_add_ps(fix1,tx);
 587             fiy1             = _mm_add_ps(fiy1,ty);
 588             fiz1             = _mm_add_ps(fiz1,tz);
 589
 590             fjx0             = _mm_add_ps(fjx0,tx);
 591             fjy0             = _mm_add_ps(fjy0,ty);
 592             fjz0             = _mm_add_ps(fjz0,tz);
 593
 594             /**************************
 595              * CALCULATE INTERACTIONS *
 596              **************************/
 597
 598             r20              = _mm_mul_ps(rsq20,rinv20);
 599             r20              = _mm_andnot_ps(dummy_mask,r20);
 600
 601             /* Compute parameters for interactions between i and j atoms */
 602             qq20             = _mm_mul_ps(iq2,jq0);
 603
 604             /* EWALD ELECTROSTATICS */
 605
 606             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 607             ewrt             = _mm_mul_ps(r20,ewtabscale);
 608             ewitab           = _mm_cvttps_epi32(ewrt);
 609             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 610             ewitab           = _mm_slli_epi32(ewitab,2);
 611             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 612             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 613             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 614             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 615             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 616             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 617             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 618             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 619             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 620
 621             /* Update potential sum for this i atom from the interaction with this j atom. */
 622             velec            = _mm_andnot_ps(dummy_mask,velec);
 623             velecsum         = _mm_add_ps(velecsum,velec);
 624
 625             fscal            = felec;
 626
 627             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 628
 629             /* Calculate temporary vectorial force */
 630             tx               = _mm_mul_ps(fscal,dx20);
 631             ty               = _mm_mul_ps(fscal,dy20);
 632             tz               = _mm_mul_ps(fscal,dz20);
 633
 634             /* Update vectorial force */
 635             fix2             = _mm_add_ps(fix2,tx);
 636             fiy2             = _mm_add_ps(fiy2,ty);
 637             fiz2             = _mm_add_ps(fiz2,tz);
 638
 639             fjx0             = _mm_add_ps(fjx0,tx);
 640             fjy0             = _mm_add_ps(fjy0,ty);
 641             fjz0             = _mm_add_ps(fjz0,tz);
 642
 643             /**************************
 644              * CALCULATE INTERACTIONS *
 645              **************************/
 646
 647             r30              = _mm_mul_ps(rsq30,rinv30);
 648             r30              = _mm_andnot_ps(dummy_mask,r30);
 649
 650             /* Compute parameters for interactions between i and j atoms */
 651             qq30             = _mm_mul_ps(iq3,jq0);
 652
 653             /* EWALD ELECTROSTATICS */
 654
 655             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 656             ewrt             = _mm_mul_ps(r30,ewtabscale);
 657             ewitab           = _mm_cvttps_epi32(ewrt);
 658             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 659             ewitab           = _mm_slli_epi32(ewitab,2);
 660             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 661             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 662             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 663             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 664             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 665             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 666             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 667             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 668             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 669
 670             /* Update potential sum for this i atom from the interaction with this j atom. */
 671             velec            = _mm_andnot_ps(dummy_mask,velec);
 672             velecsum         = _mm_add_ps(velecsum,velec);
 673
 674             fscal            = felec;
 675
 676             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 677
 678             /* Calculate temporary vectorial force */
 679             tx               = _mm_mul_ps(fscal,dx30);
 680             ty               = _mm_mul_ps(fscal,dy30);
 681             tz               = _mm_mul_ps(fscal,dz30);
 682
 683             /* Update vectorial force */
 684             fix3             = _mm_add_ps(fix3,tx);
 685             fiy3             = _mm_add_ps(fiy3,ty);
 686             fiz3             = _mm_add_ps(fiz3,tz);
 687
 688             fjx0             = _mm_add_ps(fjx0,tx);
 689             fjy0             = _mm_add_ps(fjy0,ty);
 690             fjz0             = _mm_add_ps(fjz0,tz);
 691
 692             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 693             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 694             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 695             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 696
 697             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 698
 699             /* Inner loop uses 158 flops */
 700         }
 701
 702         /* End of innermost loop */
 703
 704         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 705                                               f+i_coord_offset,fshift+i_shift_offset);
 706
 707         ggid                        = gid[iidx];
 708         /* Update potential energies */
 709         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 710         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 711
 712         /* Increment number of inner iterations */
 713         inneriter                  += j_index_end - j_index_start;
 714
 715         /* Outer loop uses 26 flops */
 716     }
 717
 718     /* Increment number of outer iterations */
 719     outeriter        += nri;
 720
 721     /* Update outer/inner flops */
 722
 723     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*158);
 724 }
 725 /*
 726  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_sse4_1_single
 727  * Electrostatics interaction: Ewald
 728  * VdW interaction:            LennardJones
 729  * Geometry:                   Water4-Particle
 730  * Calculate force/pot:        Force
 731  */
 732 void
 733 nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_sse4_1_single
 734                     (t_nblist                    * gmx_restrict       nlist,
 735                      rvec                        * gmx_restrict          xx,
 736                      rvec                        * gmx_restrict          ff,
 737                      t_forcerec                  * gmx_restrict          fr,
 738                      t_mdatoms                   * gmx_restrict     mdatoms,
 739                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 740                      t_nrnb                      * gmx_restrict        nrnb)
 741 {
 742     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 743      * just 0 for non-waters.
 744      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 745      * jnr indices corresponding to data put in the four positions in the SIMD register.
 746      */
 747     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 748     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 749     int              jnrA,jnrB,jnrC,jnrD;
 750     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 751     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 752     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 753     real             rcutoff_scalar;
 754     real             *shiftvec,*fshift,*x,*f;
 755     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 756     real             scratch[4*DIM];
 757     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 758     int              vdwioffset0;
 759     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 760     int              vdwioffset1;
 761     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 762     int              vdwioffset2;
 763     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 764     int              vdwioffset3;
 765     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 766     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 767     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 768     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 769     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 770     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 771     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 772     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 773     real             *charge;
 774     int              nvdwtype;
 775     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 776     int              *vdwtype;
 777     real             *vdwparam;
 778     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 779     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 780     __m128i          ewitab;
 781     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 782     real             *ewtab;
 783     __m128           dummy_mask,cutoff_mask;
 784     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 785     __m128           one     = _mm_set1_ps(1.0);
 786     __m128           two     = _mm_set1_ps(2.0);
 787     x                = xx[0];
 788     f                = ff[0];
 789
 790     nri              = nlist->nri;
 791     iinr             = nlist->iinr;
 792     jindex           = nlist->jindex;
 793     jjnr             = nlist->jjnr;
 794     shiftidx         = nlist->shift;
 795     gid              = nlist->gid;
 796     shiftvec         = fr->shift_vec[0];
 797     fshift           = fr->fshift[0];
 798     facel            = _mm_set1_ps(fr->epsfac);
 799     charge           = mdatoms->chargeA;
 800     nvdwtype         = fr->ntype;
 801     vdwparam         = fr->nbfp;
 802     vdwtype          = mdatoms->typeA;
 803
 804     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 805     ewtab            = fr->ic->tabq_coul_F;
 806     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 807     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 808
 809     /* Setup water-specific parameters */
 810     inr              = nlist->iinr[0];
 811     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 812     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 813     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 814     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 815
 816     /* Avoid stupid compiler warnings */
 817     jnrA = jnrB = jnrC = jnrD = 0;
 818     j_coord_offsetA = 0;
 819     j_coord_offsetB = 0;
 820     j_coord_offsetC = 0;
 821     j_coord_offsetD = 0;
 822
 823     outeriter        = 0;
 824     inneriter        = 0;
 825
 826     for(iidx=0;iidx<4*DIM;iidx++)
 827     {
 828         scratch[iidx] = 0.0;
 829     }
 830
 831     /* Start outer loop over neighborlists */
 832     for(iidx=0; iidx<nri; iidx++)
 833     {
 834         /* Load shift vector for this list */
 835         i_shift_offset   = DIM*shiftidx[iidx];
 836
 837         /* Load limits for loop over neighbors */
 838         j_index_start    = jindex[iidx];
 839         j_index_end      = jindex[iidx+1];
 840
 841         /* Get outer coordinate index */
 842         inr              = iinr[iidx];
 843         i_coord_offset   = DIM*inr;
 844
 845         /* Load i particle coords and add shift vector */
 846         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 847                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 848
 849         fix0             = _mm_setzero_ps();
 850         fiy0             = _mm_setzero_ps();
 851         fiz0             = _mm_setzero_ps();
 852         fix1             = _mm_setzero_ps();
 853         fiy1             = _mm_setzero_ps();
 854         fiz1             = _mm_setzero_ps();
 855         fix2             = _mm_setzero_ps();
 856         fiy2             = _mm_setzero_ps();
 857         fiz2             = _mm_setzero_ps();
 858         fix3             = _mm_setzero_ps();
 859         fiy3             = _mm_setzero_ps();
 860         fiz3             = _mm_setzero_ps();
 861
 862         /* Start inner kernel loop */
 863         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 864         {
 865
 866             /* Get j neighbor index, and coordinate index */
 867             jnrA             = jjnr[jidx];
 868             jnrB             = jjnr[jidx+1];
 869             jnrC             = jjnr[jidx+2];
 870             jnrD             = jjnr[jidx+3];
 871             j_coord_offsetA  = DIM*jnrA;
 872             j_coord_offsetB  = DIM*jnrB;
 873             j_coord_offsetC  = DIM*jnrC;
 874             j_coord_offsetD  = DIM*jnrD;
 875
 876             /* load j atom coordinates */
 877             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 878                                               x+j_coord_offsetC,x+j_coord_offsetD,
 879                                               &jx0,&jy0,&jz0);
 880
 881             /* Calculate displacement vector */
 882             dx00             = _mm_sub_ps(ix0,jx0);
 883             dy00             = _mm_sub_ps(iy0,jy0);
 884             dz00             = _mm_sub_ps(iz0,jz0);
 885             dx10             = _mm_sub_ps(ix1,jx0);
 886             dy10             = _mm_sub_ps(iy1,jy0);
 887             dz10             = _mm_sub_ps(iz1,jz0);
 888             dx20             = _mm_sub_ps(ix2,jx0);
 889             dy20             = _mm_sub_ps(iy2,jy0);
 890             dz20             = _mm_sub_ps(iz2,jz0);
 891             dx30             = _mm_sub_ps(ix3,jx0);
 892             dy30             = _mm_sub_ps(iy3,jy0);
 893             dz30             = _mm_sub_ps(iz3,jz0);
 894
 895             /* Calculate squared distance and things based on it */
 896             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 897             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 898             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 899             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 900
 901             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 902             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 903             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 904
 905             rinvsq00         = gmx_mm_inv_ps(rsq00);
 906             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 907             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 908             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 909
 910             /* Load parameters for j particles */
 911             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 912                                                               charge+jnrC+0,charge+jnrD+0);
 913             vdwjidx0A        = 2*vdwtype[jnrA+0];
 914             vdwjidx0B        = 2*vdwtype[jnrB+0];
 915             vdwjidx0C        = 2*vdwtype[jnrC+0];
 916             vdwjidx0D        = 2*vdwtype[jnrD+0];
 917
 918             fjx0             = _mm_setzero_ps();
 919             fjy0             = _mm_setzero_ps();
 920             fjz0             = _mm_setzero_ps();
 921
 922             /**************************
 923              * CALCULATE INTERACTIONS *
 924              **************************/
 925
 926             /* Compute parameters for interactions between i and j atoms */
 927             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 928                                          vdwparam+vdwioffset0+vdwjidx0B,
 929                                          vdwparam+vdwioffset0+vdwjidx0C,
 930                                          vdwparam+vdwioffset0+vdwjidx0D,
 931                                          &c6_00,&c12_00);
 932
 933             /* LENNARD-JONES DISPERSION/REPULSION */
 934
 935             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 936             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 937
 938             fscal            = fvdw;
 939
 940             /* Calculate temporary vectorial force */
 941             tx               = _mm_mul_ps(fscal,dx00);
 942             ty               = _mm_mul_ps(fscal,dy00);
 943             tz               = _mm_mul_ps(fscal,dz00);
 944
 945             /* Update vectorial force */
 946             fix0             = _mm_add_ps(fix0,tx);
 947             fiy0             = _mm_add_ps(fiy0,ty);
 948             fiz0             = _mm_add_ps(fiz0,tz);
 949
 950             fjx0             = _mm_add_ps(fjx0,tx);
 951             fjy0             = _mm_add_ps(fjy0,ty);
 952             fjz0             = _mm_add_ps(fjz0,tz);
 953
 954             /**************************
 955              * CALCULATE INTERACTIONS *
 956              **************************/
 957
 958             r10              = _mm_mul_ps(rsq10,rinv10);
 959
 960             /* Compute parameters for interactions between i and j atoms */
 961             qq10             = _mm_mul_ps(iq1,jq0);
 962
 963             /* EWALD ELECTROSTATICS */
 964
 965             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 966             ewrt             = _mm_mul_ps(r10,ewtabscale);
 967             ewitab           = _mm_cvttps_epi32(ewrt);
 968             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 969             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 970                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 971                                          &ewtabF,&ewtabFn);
 972             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 973             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 974
 975             fscal            = felec;
 976
 977             /* Calculate temporary vectorial force */
 978             tx               = _mm_mul_ps(fscal,dx10);
 979             ty               = _mm_mul_ps(fscal,dy10);
 980             tz               = _mm_mul_ps(fscal,dz10);
 981
 982             /* Update vectorial force */
 983             fix1             = _mm_add_ps(fix1,tx);
 984             fiy1             = _mm_add_ps(fiy1,ty);
 985             fiz1             = _mm_add_ps(fiz1,tz);
 986
 987             fjx0             = _mm_add_ps(fjx0,tx);
 988             fjy0             = _mm_add_ps(fjy0,ty);
 989             fjz0             = _mm_add_ps(fjz0,tz);
 990
 991             /**************************
 992              * CALCULATE INTERACTIONS *
 993              **************************/
 994
 995             r20              = _mm_mul_ps(rsq20,rinv20);
 996
 997             /* Compute parameters for interactions between i and j atoms */
 998             qq20             = _mm_mul_ps(iq2,jq0);
 999
1000             /* EWALD ELECTROSTATICS */
1001
1002             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1003             ewrt             = _mm_mul_ps(r20,ewtabscale);
1004             ewitab           = _mm_cvttps_epi32(ewrt);
1005             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1006             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1007                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1008                                          &ewtabF,&ewtabFn);
1009             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1010             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1011
1012             fscal            = felec;
1013
1014             /* Calculate temporary vectorial force */
1015             tx               = _mm_mul_ps(fscal,dx20);
1016             ty               = _mm_mul_ps(fscal,dy20);
1017             tz               = _mm_mul_ps(fscal,dz20);
1018
1019             /* Update vectorial force */
1020             fix2             = _mm_add_ps(fix2,tx);
1021             fiy2             = _mm_add_ps(fiy2,ty);
1022             fiz2             = _mm_add_ps(fiz2,tz);
1023
1024             fjx0             = _mm_add_ps(fjx0,tx);
1025             fjy0             = _mm_add_ps(fjy0,ty);
1026             fjz0             = _mm_add_ps(fjz0,tz);
1027
1028             /**************************
1029              * CALCULATE INTERACTIONS *
1030              **************************/
1031
1032             r30              = _mm_mul_ps(rsq30,rinv30);
1033
1034             /* Compute parameters for interactions between i and j atoms */
1035             qq30             = _mm_mul_ps(iq3,jq0);
1036
1037             /* EWALD ELECTROSTATICS */
1038
1039             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1040             ewrt             = _mm_mul_ps(r30,ewtabscale);
1041             ewitab           = _mm_cvttps_epi32(ewrt);
1042             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1043             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1044                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1045                                          &ewtabF,&ewtabFn);
1046             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1047             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1048
1049             fscal            = felec;
1050
1051             /* Calculate temporary vectorial force */
1052             tx               = _mm_mul_ps(fscal,dx30);
1053             ty               = _mm_mul_ps(fscal,dy30);
1054             tz               = _mm_mul_ps(fscal,dz30);
1055
1056             /* Update vectorial force */
1057             fix3             = _mm_add_ps(fix3,tx);
1058             fiy3             = _mm_add_ps(fiy3,ty);
1059             fiz3             = _mm_add_ps(fiz3,tz);
1060
1061             fjx0             = _mm_add_ps(fjx0,tx);
1062             fjy0             = _mm_add_ps(fjy0,ty);
1063             fjz0             = _mm_add_ps(fjz0,tz);
1064
1065             fjptrA             = f+j_coord_offsetA;
1066             fjptrB             = f+j_coord_offsetB;
1067             fjptrC             = f+j_coord_offsetC;
1068             fjptrD             = f+j_coord_offsetD;
1069
1070             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1071
1072             /* Inner loop uses 135 flops */
1073         }
1074
1075         if(jidx<j_index_end)
1076         {
1077
1078             /* Get j neighbor index, and coordinate index */
1079             jnrlistA         = jjnr[jidx];
1080             jnrlistB         = jjnr[jidx+1];
1081             jnrlistC         = jjnr[jidx+2];
1082             jnrlistD         = jjnr[jidx+3];
1083             /* Sign of each element will be negative for non-real atoms.
1084              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1085              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1086              */
1087             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1088             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1089             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1090             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1091             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1092             j_coord_offsetA  = DIM*jnrA;
1093             j_coord_offsetB  = DIM*jnrB;
1094             j_coord_offsetC  = DIM*jnrC;
1095             j_coord_offsetD  = DIM*jnrD;
1096
1097             /* load j atom coordinates */
1098             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1099                                               x+j_coord_offsetC,x+j_coord_offsetD,
1100                                               &jx0,&jy0,&jz0);
1101
1102             /* Calculate displacement vector */
1103             dx00             = _mm_sub_ps(ix0,jx0);
1104             dy00             = _mm_sub_ps(iy0,jy0);
1105             dz00             = _mm_sub_ps(iz0,jz0);
1106             dx10             = _mm_sub_ps(ix1,jx0);
1107             dy10             = _mm_sub_ps(iy1,jy0);
1108             dz10             = _mm_sub_ps(iz1,jz0);
1109             dx20             = _mm_sub_ps(ix2,jx0);
1110             dy20             = _mm_sub_ps(iy2,jy0);
1111             dz20             = _mm_sub_ps(iz2,jz0);
1112             dx30             = _mm_sub_ps(ix3,jx0);
1113             dy30             = _mm_sub_ps(iy3,jy0);
1114             dz30             = _mm_sub_ps(iz3,jz0);
1115
1116             /* Calculate squared distance and things based on it */
1117             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1118             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1119             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1120             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1121
1122             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1123             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1124             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1125
1126             rinvsq00         = gmx_mm_inv_ps(rsq00);
1127             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1128             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1129             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1130
1131             /* Load parameters for j particles */
1132             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1133                                                               charge+jnrC+0,charge+jnrD+0);
1134             vdwjidx0A        = 2*vdwtype[jnrA+0];
1135             vdwjidx0B        = 2*vdwtype[jnrB+0];
1136             vdwjidx0C        = 2*vdwtype[jnrC+0];
1137             vdwjidx0D        = 2*vdwtype[jnrD+0];
1138
1139             fjx0             = _mm_setzero_ps();
1140             fjy0             = _mm_setzero_ps();
1141             fjz0             = _mm_setzero_ps();
1142
1143             /**************************
1144              * CALCULATE INTERACTIONS *
1145              **************************/
1146
1147             /* Compute parameters for interactions between i and j atoms */
1148             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1149                                          vdwparam+vdwioffset0+vdwjidx0B,
1150                                          vdwparam+vdwioffset0+vdwjidx0C,
1151                                          vdwparam+vdwioffset0+vdwjidx0D,
1152                                          &c6_00,&c12_00);
1153
1154             /* LENNARD-JONES DISPERSION/REPULSION */
1155
1156             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1157             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1158
1159             fscal            = fvdw;
1160
1161             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1162
1163             /* Calculate temporary vectorial force */
1164             tx               = _mm_mul_ps(fscal,dx00);
1165             ty               = _mm_mul_ps(fscal,dy00);
1166             tz               = _mm_mul_ps(fscal,dz00);
1167
1168             /* Update vectorial force */
1169             fix0             = _mm_add_ps(fix0,tx);
1170             fiy0             = _mm_add_ps(fiy0,ty);
1171             fiz0             = _mm_add_ps(fiz0,tz);
1172
1173             fjx0             = _mm_add_ps(fjx0,tx);
1174             fjy0             = _mm_add_ps(fjy0,ty);
1175             fjz0             = _mm_add_ps(fjz0,tz);
1176
1177             /**************************
1178              * CALCULATE INTERACTIONS *
1179              **************************/
1180
1181             r10              = _mm_mul_ps(rsq10,rinv10);
1182             r10              = _mm_andnot_ps(dummy_mask,r10);
1183
1184             /* Compute parameters for interactions between i and j atoms */
1185             qq10             = _mm_mul_ps(iq1,jq0);
1186
1187             /* EWALD ELECTROSTATICS */
1188
1189             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1190             ewrt             = _mm_mul_ps(r10,ewtabscale);
1191             ewitab           = _mm_cvttps_epi32(ewrt);
1192             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1193             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1194                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1195                                          &ewtabF,&ewtabFn);
1196             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1197             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1198
1199             fscal            = felec;
1200
1201             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1202
1203             /* Calculate temporary vectorial force */
1204             tx               = _mm_mul_ps(fscal,dx10);
1205             ty               = _mm_mul_ps(fscal,dy10);
1206             tz               = _mm_mul_ps(fscal,dz10);
1207
1208             /* Update vectorial force */
1209             fix1             = _mm_add_ps(fix1,tx);
1210             fiy1             = _mm_add_ps(fiy1,ty);
1211             fiz1             = _mm_add_ps(fiz1,tz);
1212
1213             fjx0             = _mm_add_ps(fjx0,tx);
1214             fjy0             = _mm_add_ps(fjy0,ty);
1215             fjz0             = _mm_add_ps(fjz0,tz);
1216
1217             /**************************
1218              * CALCULATE INTERACTIONS *
1219              **************************/
1220
1221             r20              = _mm_mul_ps(rsq20,rinv20);
1222             r20              = _mm_andnot_ps(dummy_mask,r20);
1223
1224             /* Compute parameters for interactions between i and j atoms */
1225             qq20             = _mm_mul_ps(iq2,jq0);
1226
1227             /* EWALD ELECTROSTATICS */
1228
1229             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1230             ewrt             = _mm_mul_ps(r20,ewtabscale);
1231             ewitab           = _mm_cvttps_epi32(ewrt);
1232             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1233             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1234                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1235                                          &ewtabF,&ewtabFn);
1236             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1237             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1238
1239             fscal            = felec;
1240
1241             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1242
1243             /* Calculate temporary vectorial force */
1244             tx               = _mm_mul_ps(fscal,dx20);
1245             ty               = _mm_mul_ps(fscal,dy20);
1246             tz               = _mm_mul_ps(fscal,dz20);
1247
1248             /* Update vectorial force */
1249             fix2             = _mm_add_ps(fix2,tx);
1250             fiy2             = _mm_add_ps(fiy2,ty);
1251             fiz2             = _mm_add_ps(fiz2,tz);
1252
1253             fjx0             = _mm_add_ps(fjx0,tx);
1254             fjy0             = _mm_add_ps(fjy0,ty);
1255             fjz0             = _mm_add_ps(fjz0,tz);
1256
1257             /**************************
1258              * CALCULATE INTERACTIONS *
1259              **************************/
1260
1261             r30              = _mm_mul_ps(rsq30,rinv30);
1262             r30              = _mm_andnot_ps(dummy_mask,r30);
1263
1264             /* Compute parameters for interactions between i and j atoms */
1265             qq30             = _mm_mul_ps(iq3,jq0);
1266
1267             /* EWALD ELECTROSTATICS */
1268
1269             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1270             ewrt             = _mm_mul_ps(r30,ewtabscale);
1271             ewitab           = _mm_cvttps_epi32(ewrt);
1272             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1273             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1274                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1275                                          &ewtabF,&ewtabFn);
1276             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1277             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1278
1279             fscal            = felec;
1280
1281             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1282
1283             /* Calculate temporary vectorial force */
1284             tx               = _mm_mul_ps(fscal,dx30);
1285             ty               = _mm_mul_ps(fscal,dy30);
1286             tz               = _mm_mul_ps(fscal,dz30);
1287
1288             /* Update vectorial force */
1289             fix3             = _mm_add_ps(fix3,tx);
1290             fiy3             = _mm_add_ps(fiy3,ty);
1291             fiz3             = _mm_add_ps(fiz3,tz);
1292
1293             fjx0             = _mm_add_ps(fjx0,tx);
1294             fjy0             = _mm_add_ps(fjy0,ty);
1295             fjz0             = _mm_add_ps(fjz0,tz);
1296
1297             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1298             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1299             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1300             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1301
1302             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1303
1304             /* Inner loop uses 138 flops */
1305         }
1306
1307         /* End of innermost loop */
1308
1309         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1310                                               f+i_coord_offset,fshift+i_shift_offset);
1311
1312         /* Increment number of inner iterations */
1313         inneriter                  += j_index_end - j_index_start;
1314
1315         /* Outer loop uses 24 flops */
1316     }
1317
1318     /* Increment number of outer iterations */
1319     outeriter        += nri;
1320
1321     /* Update outer/inner flops */
1322
1323     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*138);
1324 }