src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEw_VdwLJEw_GeomW3P1_sse2_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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_VdwLJEw_GeomW3P1_VF_sse2_single
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LJEwald
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_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              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  89     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 100     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 101     __m128           c6grid_00;
 102     __m128           c6grid_10;
 103     __m128           c6grid_20;
 104     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 105     real             *vdwgridparam;
 106     __m128           one_half = _mm_set1_ps(0.5);
 107     __m128           minus_one = _mm_set1_ps(-1.0);
 108     __m128i          ewitab;
 109     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 110     real             *ewtab;
 111     __m128           dummy_mask,cutoff_mask;
 112     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 113     __m128           one     = _mm_set1_ps(1.0);
 114     __m128           two     = _mm_set1_ps(2.0);
 115     x                = xx[0];
 116     f                = ff[0];
 117
 118     nri              = nlist->nri;
 119     iinr             = nlist->iinr;
 120     jindex           = nlist->jindex;
 121     jjnr             = nlist->jjnr;
 122     shiftidx         = nlist->shift;
 123     gid              = nlist->gid;
 124     shiftvec         = fr->shift_vec[0];
 125     fshift           = fr->fshift[0];
 126     facel            = _mm_set1_ps(fr->ic->epsfac);
 127     charge           = mdatoms->chargeA;
 128     nvdwtype         = fr->ntype;
 129     vdwparam         = fr->nbfp;
 130     vdwtype          = mdatoms->typeA;
 131     vdwgridparam     = fr->ljpme_c6grid;
 132     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 133     ewclj            = _mm_set1_ps(fr->ic->ewaldcoeff_lj);
 134     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 135
 136     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 137     ewtab            = fr->ic->tabq_coul_FDV0;
 138     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 139     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 140
 141     /* Setup water-specific parameters */
 142     inr              = nlist->iinr[0];
 143     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 144     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 145     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 146     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 147
 148     /* Avoid stupid compiler warnings */
 149     jnrA = jnrB = jnrC = jnrD = 0;
 150     j_coord_offsetA = 0;
 151     j_coord_offsetB = 0;
 152     j_coord_offsetC = 0;
 153     j_coord_offsetD = 0;
 154
 155     outeriter        = 0;
 156     inneriter        = 0;
 157
 158     for(iidx=0;iidx<4*DIM;iidx++)
 159     {
 160         scratch[iidx] = 0.0;
 161     }
 162
 163     /* Start outer loop over neighborlists */
 164     for(iidx=0; iidx<nri; iidx++)
 165     {
 166         /* Load shift vector for this list */
 167         i_shift_offset   = DIM*shiftidx[iidx];
 168
 169         /* Load limits for loop over neighbors */
 170         j_index_start    = jindex[iidx];
 171         j_index_end      = jindex[iidx+1];
 172
 173         /* Get outer coordinate index */
 174         inr              = iinr[iidx];
 175         i_coord_offset   = DIM*inr;
 176
 177         /* Load i particle coords and add shift vector */
 178         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 179                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 180
 181         fix0             = _mm_setzero_ps();
 182         fiy0             = _mm_setzero_ps();
 183         fiz0             = _mm_setzero_ps();
 184         fix1             = _mm_setzero_ps();
 185         fiy1             = _mm_setzero_ps();
 186         fiz1             = _mm_setzero_ps();
 187         fix2             = _mm_setzero_ps();
 188         fiy2             = _mm_setzero_ps();
 189         fiz2             = _mm_setzero_ps();
 190
 191         /* Reset potential sums */
 192         velecsum         = _mm_setzero_ps();
 193         vvdwsum          = _mm_setzero_ps();
 194
 195         /* Start inner kernel loop */
 196         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 197         {
 198
 199             /* Get j neighbor index, and coordinate index */
 200             jnrA             = jjnr[jidx];
 201             jnrB             = jjnr[jidx+1];
 202             jnrC             = jjnr[jidx+2];
 203             jnrD             = jjnr[jidx+3];
 204             j_coord_offsetA  = DIM*jnrA;
 205             j_coord_offsetB  = DIM*jnrB;
 206             j_coord_offsetC  = DIM*jnrC;
 207             j_coord_offsetD  = DIM*jnrD;
 208
 209             /* load j atom coordinates */
 210             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 211                                               x+j_coord_offsetC,x+j_coord_offsetD,
 212                                               &jx0,&jy0,&jz0);
 213
 214             /* Calculate displacement vector */
 215             dx00             = _mm_sub_ps(ix0,jx0);
 216             dy00             = _mm_sub_ps(iy0,jy0);
 217             dz00             = _mm_sub_ps(iz0,jz0);
 218             dx10             = _mm_sub_ps(ix1,jx0);
 219             dy10             = _mm_sub_ps(iy1,jy0);
 220             dz10             = _mm_sub_ps(iz1,jz0);
 221             dx20             = _mm_sub_ps(ix2,jx0);
 222             dy20             = _mm_sub_ps(iy2,jy0);
 223             dz20             = _mm_sub_ps(iz2,jz0);
 224
 225             /* Calculate squared distance and things based on it */
 226             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 227             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 228             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 229
 230             rinv00           = sse2_invsqrt_f(rsq00);
 231             rinv10           = sse2_invsqrt_f(rsq10);
 232             rinv20           = sse2_invsqrt_f(rsq20);
 233
 234             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 235             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 236             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 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             r00              = _mm_mul_ps(rsq00,rinv00);
 255
 256             /* Compute parameters for interactions between i and j atoms */
 257             qq00             = _mm_mul_ps(iq0,jq0);
 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             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 264                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 265                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 266                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 267
 268             /* EWALD ELECTROSTATICS */
 269
 270             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 271             ewrt             = _mm_mul_ps(r00,ewtabscale);
 272             ewitab           = _mm_cvttps_epi32(ewrt);
 273             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 274             ewitab           = _mm_slli_epi32(ewitab,2);
 275             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 276             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 277             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 278             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 279             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 280             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 281             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 282             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 283             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 284
 285             /* Analytical LJ-PME */
 286             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 287             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 288             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 289             exponent         = sse2_exp_f(ewcljrsq);
 290             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 291             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 292             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 293             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 294             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 295             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 296             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 297             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,_mm_sub_ps(vvdw6,_mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6)))),rinvsq00);
 298
 299             /* Update potential sum for this i atom from the interaction with this j atom. */
 300             velecsum         = _mm_add_ps(velecsum,velec);
 301             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 302
 303             fscal            = _mm_add_ps(felec,fvdw);
 304
 305             /* Calculate temporary vectorial force */
 306             tx               = _mm_mul_ps(fscal,dx00);
 307             ty               = _mm_mul_ps(fscal,dy00);
 308             tz               = _mm_mul_ps(fscal,dz00);
 309
 310             /* Update vectorial force */
 311             fix0             = _mm_add_ps(fix0,tx);
 312             fiy0             = _mm_add_ps(fiy0,ty);
 313             fiz0             = _mm_add_ps(fiz0,tz);
 314
 315             fjx0             = _mm_add_ps(fjx0,tx);
 316             fjy0             = _mm_add_ps(fjy0,ty);
 317             fjz0             = _mm_add_ps(fjz0,tz);
 318
 319             /**************************
 320              * CALCULATE INTERACTIONS *
 321              **************************/
 322
 323             r10              = _mm_mul_ps(rsq10,rinv10);
 324
 325             /* Compute parameters for interactions between i and j atoms */
 326             qq10             = _mm_mul_ps(iq1,jq0);
 327
 328             /* EWALD ELECTROSTATICS */
 329
 330             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 331             ewrt             = _mm_mul_ps(r10,ewtabscale);
 332             ewitab           = _mm_cvttps_epi32(ewrt);
 333             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 334             ewitab           = _mm_slli_epi32(ewitab,2);
 335             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 336             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 337             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 338             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 339             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 340             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 341             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 342             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 343             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 344
 345             /* Update potential sum for this i atom from the interaction with this j atom. */
 346             velecsum         = _mm_add_ps(velecsum,velec);
 347
 348             fscal            = felec;
 349
 350             /* Calculate temporary vectorial force */
 351             tx               = _mm_mul_ps(fscal,dx10);
 352             ty               = _mm_mul_ps(fscal,dy10);
 353             tz               = _mm_mul_ps(fscal,dz10);
 354
 355             /* Update vectorial force */
 356             fix1             = _mm_add_ps(fix1,tx);
 357             fiy1             = _mm_add_ps(fiy1,ty);
 358             fiz1             = _mm_add_ps(fiz1,tz);
 359
 360             fjx0             = _mm_add_ps(fjx0,tx);
 361             fjy0             = _mm_add_ps(fjy0,ty);
 362             fjz0             = _mm_add_ps(fjz0,tz);
 363
 364             /**************************
 365              * CALCULATE INTERACTIONS *
 366              **************************/
 367
 368             r20              = _mm_mul_ps(rsq20,rinv20);
 369
 370             /* Compute parameters for interactions between i and j atoms */
 371             qq20             = _mm_mul_ps(iq2,jq0);
 372
 373             /* EWALD ELECTROSTATICS */
 374
 375             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 376             ewrt             = _mm_mul_ps(r20,ewtabscale);
 377             ewitab           = _mm_cvttps_epi32(ewrt);
 378             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 379             ewitab           = _mm_slli_epi32(ewitab,2);
 380             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 381             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 382             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 383             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 384             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 385             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 386             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 387             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 388             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 389
 390             /* Update potential sum for this i atom from the interaction with this j atom. */
 391             velecsum         = _mm_add_ps(velecsum,velec);
 392
 393             fscal            = felec;
 394
 395             /* Calculate temporary vectorial force */
 396             tx               = _mm_mul_ps(fscal,dx20);
 397             ty               = _mm_mul_ps(fscal,dy20);
 398             tz               = _mm_mul_ps(fscal,dz20);
 399
 400             /* Update vectorial force */
 401             fix2             = _mm_add_ps(fix2,tx);
 402             fiy2             = _mm_add_ps(fiy2,ty);
 403             fiz2             = _mm_add_ps(fiz2,tz);
 404
 405             fjx0             = _mm_add_ps(fjx0,tx);
 406             fjy0             = _mm_add_ps(fjy0,ty);
 407             fjz0             = _mm_add_ps(fjz0,tz);
 408
 409             fjptrA             = f+j_coord_offsetA;
 410             fjptrB             = f+j_coord_offsetB;
 411             fjptrC             = f+j_coord_offsetC;
 412             fjptrD             = f+j_coord_offsetD;
 413
 414             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 415
 416             /* Inner loop uses 151 flops */
 417         }
 418
 419         if(jidx<j_index_end)
 420         {
 421
 422             /* Get j neighbor index, and coordinate index */
 423             jnrlistA         = jjnr[jidx];
 424             jnrlistB         = jjnr[jidx+1];
 425             jnrlistC         = jjnr[jidx+2];
 426             jnrlistD         = jjnr[jidx+3];
 427             /* Sign of each element will be negative for non-real atoms.
 428              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 429              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 430              */
 431             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 432             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 433             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 434             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 435             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 436             j_coord_offsetA  = DIM*jnrA;
 437             j_coord_offsetB  = DIM*jnrB;
 438             j_coord_offsetC  = DIM*jnrC;
 439             j_coord_offsetD  = DIM*jnrD;
 440
 441             /* load j atom coordinates */
 442             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 443                                               x+j_coord_offsetC,x+j_coord_offsetD,
 444                                               &jx0,&jy0,&jz0);
 445
 446             /* Calculate displacement vector */
 447             dx00             = _mm_sub_ps(ix0,jx0);
 448             dy00             = _mm_sub_ps(iy0,jy0);
 449             dz00             = _mm_sub_ps(iz0,jz0);
 450             dx10             = _mm_sub_ps(ix1,jx0);
 451             dy10             = _mm_sub_ps(iy1,jy0);
 452             dz10             = _mm_sub_ps(iz1,jz0);
 453             dx20             = _mm_sub_ps(ix2,jx0);
 454             dy20             = _mm_sub_ps(iy2,jy0);
 455             dz20             = _mm_sub_ps(iz2,jz0);
 456
 457             /* Calculate squared distance and things based on it */
 458             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 459             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 460             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 461
 462             rinv00           = sse2_invsqrt_f(rsq00);
 463             rinv10           = sse2_invsqrt_f(rsq10);
 464             rinv20           = sse2_invsqrt_f(rsq20);
 465
 466             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 467             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 468             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 469
 470             /* Load parameters for j particles */
 471             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 472                                                               charge+jnrC+0,charge+jnrD+0);
 473             vdwjidx0A        = 2*vdwtype[jnrA+0];
 474             vdwjidx0B        = 2*vdwtype[jnrB+0];
 475             vdwjidx0C        = 2*vdwtype[jnrC+0];
 476             vdwjidx0D        = 2*vdwtype[jnrD+0];
 477
 478             fjx0             = _mm_setzero_ps();
 479             fjy0             = _mm_setzero_ps();
 480             fjz0             = _mm_setzero_ps();
 481
 482             /**************************
 483              * CALCULATE INTERACTIONS *
 484              **************************/
 485
 486             r00              = _mm_mul_ps(rsq00,rinv00);
 487             r00              = _mm_andnot_ps(dummy_mask,r00);
 488
 489             /* Compute parameters for interactions between i and j atoms */
 490             qq00             = _mm_mul_ps(iq0,jq0);
 491             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 492                                          vdwparam+vdwioffset0+vdwjidx0B,
 493                                          vdwparam+vdwioffset0+vdwjidx0C,
 494                                          vdwparam+vdwioffset0+vdwjidx0D,
 495                                          &c6_00,&c12_00);
 496             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 497                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 498                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 499                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 500
 501             /* EWALD ELECTROSTATICS */
 502
 503             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 504             ewrt             = _mm_mul_ps(r00,ewtabscale);
 505             ewitab           = _mm_cvttps_epi32(ewrt);
 506             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 507             ewitab           = _mm_slli_epi32(ewitab,2);
 508             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 509             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 510             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 511             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 512             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 513             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 514             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 515             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 516             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 517
 518             /* Analytical LJ-PME */
 519             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 520             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 521             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 522             exponent         = sse2_exp_f(ewcljrsq);
 523             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 524             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 525             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 526             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 527             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 528             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 529             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 530             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,_mm_sub_ps(vvdw6,_mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6)))),rinvsq00);
 531
 532             /* Update potential sum for this i atom from the interaction with this j atom. */
 533             velec            = _mm_andnot_ps(dummy_mask,velec);
 534             velecsum         = _mm_add_ps(velecsum,velec);
 535             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 536             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 537
 538             fscal            = _mm_add_ps(felec,fvdw);
 539
 540             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 541
 542             /* Calculate temporary vectorial force */
 543             tx               = _mm_mul_ps(fscal,dx00);
 544             ty               = _mm_mul_ps(fscal,dy00);
 545             tz               = _mm_mul_ps(fscal,dz00);
 546
 547             /* Update vectorial force */
 548             fix0             = _mm_add_ps(fix0,tx);
 549             fiy0             = _mm_add_ps(fiy0,ty);
 550             fiz0             = _mm_add_ps(fiz0,tz);
 551
 552             fjx0             = _mm_add_ps(fjx0,tx);
 553             fjy0             = _mm_add_ps(fjy0,ty);
 554             fjz0             = _mm_add_ps(fjz0,tz);
 555
 556             /**************************
 557              * CALCULATE INTERACTIONS *
 558              **************************/
 559
 560             r10              = _mm_mul_ps(rsq10,rinv10);
 561             r10              = _mm_andnot_ps(dummy_mask,r10);
 562
 563             /* Compute parameters for interactions between i and j atoms */
 564             qq10             = _mm_mul_ps(iq1,jq0);
 565
 566             /* EWALD ELECTROSTATICS */
 567
 568             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 569             ewrt             = _mm_mul_ps(r10,ewtabscale);
 570             ewitab           = _mm_cvttps_epi32(ewrt);
 571             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 572             ewitab           = _mm_slli_epi32(ewitab,2);
 573             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 574             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 575             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 576             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 577             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 578             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 579             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 580             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 581             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 582
 583             /* Update potential sum for this i atom from the interaction with this j atom. */
 584             velec            = _mm_andnot_ps(dummy_mask,velec);
 585             velecsum         = _mm_add_ps(velecsum,velec);
 586
 587             fscal            = felec;
 588
 589             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 590
 591             /* Calculate temporary vectorial force */
 592             tx               = _mm_mul_ps(fscal,dx10);
 593             ty               = _mm_mul_ps(fscal,dy10);
 594             tz               = _mm_mul_ps(fscal,dz10);
 595
 596             /* Update vectorial force */
 597             fix1             = _mm_add_ps(fix1,tx);
 598             fiy1             = _mm_add_ps(fiy1,ty);
 599             fiz1             = _mm_add_ps(fiz1,tz);
 600
 601             fjx0             = _mm_add_ps(fjx0,tx);
 602             fjy0             = _mm_add_ps(fjy0,ty);
 603             fjz0             = _mm_add_ps(fjz0,tz);
 604
 605             /**************************
 606              * CALCULATE INTERACTIONS *
 607              **************************/
 608
 609             r20              = _mm_mul_ps(rsq20,rinv20);
 610             r20              = _mm_andnot_ps(dummy_mask,r20);
 611
 612             /* Compute parameters for interactions between i and j atoms */
 613             qq20             = _mm_mul_ps(iq2,jq0);
 614
 615             /* EWALD ELECTROSTATICS */
 616
 617             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 618             ewrt             = _mm_mul_ps(r20,ewtabscale);
 619             ewitab           = _mm_cvttps_epi32(ewrt);
 620             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 621             ewitab           = _mm_slli_epi32(ewitab,2);
 622             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 623             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 624             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 625             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 626             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 627             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 628             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 629             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 630             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 631
 632             /* Update potential sum for this i atom from the interaction with this j atom. */
 633             velec            = _mm_andnot_ps(dummy_mask,velec);
 634             velecsum         = _mm_add_ps(velecsum,velec);
 635
 636             fscal            = felec;
 637
 638             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 639
 640             /* Calculate temporary vectorial force */
 641             tx               = _mm_mul_ps(fscal,dx20);
 642             ty               = _mm_mul_ps(fscal,dy20);
 643             tz               = _mm_mul_ps(fscal,dz20);
 644
 645             /* Update vectorial force */
 646             fix2             = _mm_add_ps(fix2,tx);
 647             fiy2             = _mm_add_ps(fiy2,ty);
 648             fiz2             = _mm_add_ps(fiz2,tz);
 649
 650             fjx0             = _mm_add_ps(fjx0,tx);
 651             fjy0             = _mm_add_ps(fjy0,ty);
 652             fjz0             = _mm_add_ps(fjz0,tz);
 653
 654             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 655             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 656             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 657             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 658
 659             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 660
 661             /* Inner loop uses 154 flops */
 662         }
 663
 664         /* End of innermost loop */
 665
 666         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 667                                               f+i_coord_offset,fshift+i_shift_offset);
 668
 669         ggid                        = gid[iidx];
 670         /* Update potential energies */
 671         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 672         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 673
 674         /* Increment number of inner iterations */
 675         inneriter                  += j_index_end - j_index_start;
 676
 677         /* Outer loop uses 20 flops */
 678     }
 679
 680     /* Increment number of outer iterations */
 681     outeriter        += nri;
 682
 683     /* Update outer/inner flops */
 684
 685     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*154);
 686 }
 687 /*
 688  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_sse2_single
 689  * Electrostatics interaction: Ewald
 690  * VdW interaction:            LJEwald
 691  * Geometry:                   Water3-Particle
 692  * Calculate force/pot:        Force
 693  */
 694 void
 695 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_sse2_single
 696                     (t_nblist                    * gmx_restrict       nlist,
 697                      rvec                        * gmx_restrict          xx,
 698                      rvec                        * gmx_restrict          ff,
 699                      struct t_forcerec           * gmx_restrict          fr,
 700                      t_mdatoms                   * gmx_restrict     mdatoms,
 701                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 702                      t_nrnb                      * gmx_restrict        nrnb)
 703 {
 704     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 705      * just 0 for non-waters.
 706      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 707      * jnr indices corresponding to data put in the four positions in the SIMD register.
 708      */
 709     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 710     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 711     int              jnrA,jnrB,jnrC,jnrD;
 712     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 713     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 714     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 715     real             rcutoff_scalar;
 716     real             *shiftvec,*fshift,*x,*f;
 717     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 718     real             scratch[4*DIM];
 719     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 720     int              vdwioffset0;
 721     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 722     int              vdwioffset1;
 723     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 724     int              vdwioffset2;
 725     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 726     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 727     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 728     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 729     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 730     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 731     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 732     real             *charge;
 733     int              nvdwtype;
 734     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 735     int              *vdwtype;
 736     real             *vdwparam;
 737     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 738     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 739     __m128           c6grid_00;
 740     __m128           c6grid_10;
 741     __m128           c6grid_20;
 742     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 743     real             *vdwgridparam;
 744     __m128           one_half = _mm_set1_ps(0.5);
 745     __m128           minus_one = _mm_set1_ps(-1.0);
 746     __m128i          ewitab;
 747     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 748     real             *ewtab;
 749     __m128           dummy_mask,cutoff_mask;
 750     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 751     __m128           one     = _mm_set1_ps(1.0);
 752     __m128           two     = _mm_set1_ps(2.0);
 753     x                = xx[0];
 754     f                = ff[0];
 755
 756     nri              = nlist->nri;
 757     iinr             = nlist->iinr;
 758     jindex           = nlist->jindex;
 759     jjnr             = nlist->jjnr;
 760     shiftidx         = nlist->shift;
 761     gid              = nlist->gid;
 762     shiftvec         = fr->shift_vec[0];
 763     fshift           = fr->fshift[0];
 764     facel            = _mm_set1_ps(fr->ic->epsfac);
 765     charge           = mdatoms->chargeA;
 766     nvdwtype         = fr->ntype;
 767     vdwparam         = fr->nbfp;
 768     vdwtype          = mdatoms->typeA;
 769     vdwgridparam     = fr->ljpme_c6grid;
 770     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 771     ewclj            = _mm_set1_ps(fr->ic->ewaldcoeff_lj);
 772     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 773
 774     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 775     ewtab            = fr->ic->tabq_coul_F;
 776     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 777     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 778
 779     /* Setup water-specific parameters */
 780     inr              = nlist->iinr[0];
 781     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 782     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 783     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 784     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 785
 786     /* Avoid stupid compiler warnings */
 787     jnrA = jnrB = jnrC = jnrD = 0;
 788     j_coord_offsetA = 0;
 789     j_coord_offsetB = 0;
 790     j_coord_offsetC = 0;
 791     j_coord_offsetD = 0;
 792
 793     outeriter        = 0;
 794     inneriter        = 0;
 795
 796     for(iidx=0;iidx<4*DIM;iidx++)
 797     {
 798         scratch[iidx] = 0.0;
 799     }
 800
 801     /* Start outer loop over neighborlists */
 802     for(iidx=0; iidx<nri; iidx++)
 803     {
 804         /* Load shift vector for this list */
 805         i_shift_offset   = DIM*shiftidx[iidx];
 806
 807         /* Load limits for loop over neighbors */
 808         j_index_start    = jindex[iidx];
 809         j_index_end      = jindex[iidx+1];
 810
 811         /* Get outer coordinate index */
 812         inr              = iinr[iidx];
 813         i_coord_offset   = DIM*inr;
 814
 815         /* Load i particle coords and add shift vector */
 816         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 817                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 818
 819         fix0             = _mm_setzero_ps();
 820         fiy0             = _mm_setzero_ps();
 821         fiz0             = _mm_setzero_ps();
 822         fix1             = _mm_setzero_ps();
 823         fiy1             = _mm_setzero_ps();
 824         fiz1             = _mm_setzero_ps();
 825         fix2             = _mm_setzero_ps();
 826         fiy2             = _mm_setzero_ps();
 827         fiz2             = _mm_setzero_ps();
 828
 829         /* Start inner kernel loop */
 830         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 831         {
 832
 833             /* Get j neighbor index, and coordinate index */
 834             jnrA             = jjnr[jidx];
 835             jnrB             = jjnr[jidx+1];
 836             jnrC             = jjnr[jidx+2];
 837             jnrD             = jjnr[jidx+3];
 838             j_coord_offsetA  = DIM*jnrA;
 839             j_coord_offsetB  = DIM*jnrB;
 840             j_coord_offsetC  = DIM*jnrC;
 841             j_coord_offsetD  = DIM*jnrD;
 842
 843             /* load j atom coordinates */
 844             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 845                                               x+j_coord_offsetC,x+j_coord_offsetD,
 846                                               &jx0,&jy0,&jz0);
 847
 848             /* Calculate displacement vector */
 849             dx00             = _mm_sub_ps(ix0,jx0);
 850             dy00             = _mm_sub_ps(iy0,jy0);
 851             dz00             = _mm_sub_ps(iz0,jz0);
 852             dx10             = _mm_sub_ps(ix1,jx0);
 853             dy10             = _mm_sub_ps(iy1,jy0);
 854             dz10             = _mm_sub_ps(iz1,jz0);
 855             dx20             = _mm_sub_ps(ix2,jx0);
 856             dy20             = _mm_sub_ps(iy2,jy0);
 857             dz20             = _mm_sub_ps(iz2,jz0);
 858
 859             /* Calculate squared distance and things based on it */
 860             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 861             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 862             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 863
 864             rinv00           = sse2_invsqrt_f(rsq00);
 865             rinv10           = sse2_invsqrt_f(rsq10);
 866             rinv20           = sse2_invsqrt_f(rsq20);
 867
 868             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 869             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 870             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 871
 872             /* Load parameters for j particles */
 873             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 874                                                               charge+jnrC+0,charge+jnrD+0);
 875             vdwjidx0A        = 2*vdwtype[jnrA+0];
 876             vdwjidx0B        = 2*vdwtype[jnrB+0];
 877             vdwjidx0C        = 2*vdwtype[jnrC+0];
 878             vdwjidx0D        = 2*vdwtype[jnrD+0];
 879
 880             fjx0             = _mm_setzero_ps();
 881             fjy0             = _mm_setzero_ps();
 882             fjz0             = _mm_setzero_ps();
 883
 884             /**************************
 885              * CALCULATE INTERACTIONS *
 886              **************************/
 887
 888             r00              = _mm_mul_ps(rsq00,rinv00);
 889
 890             /* Compute parameters for interactions between i and j atoms */
 891             qq00             = _mm_mul_ps(iq0,jq0);
 892             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 893                                          vdwparam+vdwioffset0+vdwjidx0B,
 894                                          vdwparam+vdwioffset0+vdwjidx0C,
 895                                          vdwparam+vdwioffset0+vdwjidx0D,
 896                                          &c6_00,&c12_00);
 897             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 898                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 899                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 900                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 901
 902             /* EWALD ELECTROSTATICS */
 903
 904             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 905             ewrt             = _mm_mul_ps(r00,ewtabscale);
 906             ewitab           = _mm_cvttps_epi32(ewrt);
 907             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 908             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 909                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 910                                          &ewtabF,&ewtabFn);
 911             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 912             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 913
 914             /* Analytical LJ-PME */
 915             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 916             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 917             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 918             exponent         = sse2_exp_f(ewcljrsq);
 919             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 920             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 921             /* f6A = 6 * C6grid * (1 - poly) */
 922             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 923             /* f6B = C6grid * exponent * beta^6 */
 924             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 925             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 926             fvdw              = _mm_mul_ps(_mm_add_ps(_mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),_mm_sub_ps(c6_00,f6A)),rinvsix),f6B),rinvsq00);
 927
 928             fscal            = _mm_add_ps(felec,fvdw);
 929
 930             /* Calculate temporary vectorial force */
 931             tx               = _mm_mul_ps(fscal,dx00);
 932             ty               = _mm_mul_ps(fscal,dy00);
 933             tz               = _mm_mul_ps(fscal,dz00);
 934
 935             /* Update vectorial force */
 936             fix0             = _mm_add_ps(fix0,tx);
 937             fiy0             = _mm_add_ps(fiy0,ty);
 938             fiz0             = _mm_add_ps(fiz0,tz);
 939
 940             fjx0             = _mm_add_ps(fjx0,tx);
 941             fjy0             = _mm_add_ps(fjy0,ty);
 942             fjz0             = _mm_add_ps(fjz0,tz);
 943
 944             /**************************
 945              * CALCULATE INTERACTIONS *
 946              **************************/
 947
 948             r10              = _mm_mul_ps(rsq10,rinv10);
 949
 950             /* Compute parameters for interactions between i and j atoms */
 951             qq10             = _mm_mul_ps(iq1,jq0);
 952
 953             /* EWALD ELECTROSTATICS */
 954
 955             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 956             ewrt             = _mm_mul_ps(r10,ewtabscale);
 957             ewitab           = _mm_cvttps_epi32(ewrt);
 958             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 959             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 960                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 961                                          &ewtabF,&ewtabFn);
 962             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 963             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 964
 965             fscal            = felec;
 966
 967             /* Calculate temporary vectorial force */
 968             tx               = _mm_mul_ps(fscal,dx10);
 969             ty               = _mm_mul_ps(fscal,dy10);
 970             tz               = _mm_mul_ps(fscal,dz10);
 971
 972             /* Update vectorial force */
 973             fix1             = _mm_add_ps(fix1,tx);
 974             fiy1             = _mm_add_ps(fiy1,ty);
 975             fiz1             = _mm_add_ps(fiz1,tz);
 976
 977             fjx0             = _mm_add_ps(fjx0,tx);
 978             fjy0             = _mm_add_ps(fjy0,ty);
 979             fjz0             = _mm_add_ps(fjz0,tz);
 980
 981             /**************************
 982              * CALCULATE INTERACTIONS *
 983              **************************/
 984
 985             r20              = _mm_mul_ps(rsq20,rinv20);
 986
 987             /* Compute parameters for interactions between i and j atoms */
 988             qq20             = _mm_mul_ps(iq2,jq0);
 989
 990             /* EWALD ELECTROSTATICS */
 991
 992             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 993             ewrt             = _mm_mul_ps(r20,ewtabscale);
 994             ewitab           = _mm_cvttps_epi32(ewrt);
 995             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 996             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 997                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 998                                          &ewtabF,&ewtabFn);
 999             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1000             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1001
1002             fscal            = felec;
1003
1004             /* Calculate temporary vectorial force */
1005             tx               = _mm_mul_ps(fscal,dx20);
1006             ty               = _mm_mul_ps(fscal,dy20);
1007             tz               = _mm_mul_ps(fscal,dz20);
1008
1009             /* Update vectorial force */
1010             fix2             = _mm_add_ps(fix2,tx);
1011             fiy2             = _mm_add_ps(fiy2,ty);
1012             fiz2             = _mm_add_ps(fiz2,tz);
1013
1014             fjx0             = _mm_add_ps(fjx0,tx);
1015             fjy0             = _mm_add_ps(fjy0,ty);
1016             fjz0             = _mm_add_ps(fjz0,tz);
1017
1018             fjptrA             = f+j_coord_offsetA;
1019             fjptrB             = f+j_coord_offsetB;
1020             fjptrC             = f+j_coord_offsetC;
1021             fjptrD             = f+j_coord_offsetD;
1022
1023             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1024
1025             /* Inner loop uses 131 flops */
1026         }
1027
1028         if(jidx<j_index_end)
1029         {
1030
1031             /* Get j neighbor index, and coordinate index */
1032             jnrlistA         = jjnr[jidx];
1033             jnrlistB         = jjnr[jidx+1];
1034             jnrlistC         = jjnr[jidx+2];
1035             jnrlistD         = jjnr[jidx+3];
1036             /* Sign of each element will be negative for non-real atoms.
1037              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1038              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1039              */
1040             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1041             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1042             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1043             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1044             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1045             j_coord_offsetA  = DIM*jnrA;
1046             j_coord_offsetB  = DIM*jnrB;
1047             j_coord_offsetC  = DIM*jnrC;
1048             j_coord_offsetD  = DIM*jnrD;
1049
1050             /* load j atom coordinates */
1051             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1052                                               x+j_coord_offsetC,x+j_coord_offsetD,
1053                                               &jx0,&jy0,&jz0);
1054
1055             /* Calculate displacement vector */
1056             dx00             = _mm_sub_ps(ix0,jx0);
1057             dy00             = _mm_sub_ps(iy0,jy0);
1058             dz00             = _mm_sub_ps(iz0,jz0);
1059             dx10             = _mm_sub_ps(ix1,jx0);
1060             dy10             = _mm_sub_ps(iy1,jy0);
1061             dz10             = _mm_sub_ps(iz1,jz0);
1062             dx20             = _mm_sub_ps(ix2,jx0);
1063             dy20             = _mm_sub_ps(iy2,jy0);
1064             dz20             = _mm_sub_ps(iz2,jz0);
1065
1066             /* Calculate squared distance and things based on it */
1067             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1068             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1069             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1070
1071             rinv00           = sse2_invsqrt_f(rsq00);
1072             rinv10           = sse2_invsqrt_f(rsq10);
1073             rinv20           = sse2_invsqrt_f(rsq20);
1074
1075             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1076             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1077             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1078
1079             /* Load parameters for j particles */
1080             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1081                                                               charge+jnrC+0,charge+jnrD+0);
1082             vdwjidx0A        = 2*vdwtype[jnrA+0];
1083             vdwjidx0B        = 2*vdwtype[jnrB+0];
1084             vdwjidx0C        = 2*vdwtype[jnrC+0];
1085             vdwjidx0D        = 2*vdwtype[jnrD+0];
1086
1087             fjx0             = _mm_setzero_ps();
1088             fjy0             = _mm_setzero_ps();
1089             fjz0             = _mm_setzero_ps();
1090
1091             /**************************
1092              * CALCULATE INTERACTIONS *
1093              **************************/
1094
1095             r00              = _mm_mul_ps(rsq00,rinv00);
1096             r00              = _mm_andnot_ps(dummy_mask,r00);
1097
1098             /* Compute parameters for interactions between i and j atoms */
1099             qq00             = _mm_mul_ps(iq0,jq0);
1100             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1101                                          vdwparam+vdwioffset0+vdwjidx0B,
1102                                          vdwparam+vdwioffset0+vdwjidx0C,
1103                                          vdwparam+vdwioffset0+vdwjidx0D,
1104                                          &c6_00,&c12_00);
1105             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
1106                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
1107                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
1108                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
1109
1110             /* EWALD ELECTROSTATICS */
1111
1112             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1113             ewrt             = _mm_mul_ps(r00,ewtabscale);
1114             ewitab           = _mm_cvttps_epi32(ewrt);
1115             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1116             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1117                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1118                                          &ewtabF,&ewtabFn);
1119             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1120             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1121
1122             /* Analytical LJ-PME */
1123             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1124             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1125             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1126             exponent         = sse2_exp_f(ewcljrsq);
1127             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1128             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1129             /* f6A = 6 * C6grid * (1 - poly) */
1130             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1131             /* f6B = C6grid * exponent * beta^6 */
1132             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1133             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1134             fvdw              = _mm_mul_ps(_mm_add_ps(_mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),_mm_sub_ps(c6_00,f6A)),rinvsix),f6B),rinvsq00);
1135
1136             fscal            = _mm_add_ps(felec,fvdw);
1137
1138             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1139
1140             /* Calculate temporary vectorial force */
1141             tx               = _mm_mul_ps(fscal,dx00);
1142             ty               = _mm_mul_ps(fscal,dy00);
1143             tz               = _mm_mul_ps(fscal,dz00);
1144
1145             /* Update vectorial force */
1146             fix0             = _mm_add_ps(fix0,tx);
1147             fiy0             = _mm_add_ps(fiy0,ty);
1148             fiz0             = _mm_add_ps(fiz0,tz);
1149
1150             fjx0             = _mm_add_ps(fjx0,tx);
1151             fjy0             = _mm_add_ps(fjy0,ty);
1152             fjz0             = _mm_add_ps(fjz0,tz);
1153
1154             /**************************
1155              * CALCULATE INTERACTIONS *
1156              **************************/
1157
1158             r10              = _mm_mul_ps(rsq10,rinv10);
1159             r10              = _mm_andnot_ps(dummy_mask,r10);
1160
1161             /* Compute parameters for interactions between i and j atoms */
1162             qq10             = _mm_mul_ps(iq1,jq0);
1163
1164             /* EWALD ELECTROSTATICS */
1165
1166             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1167             ewrt             = _mm_mul_ps(r10,ewtabscale);
1168             ewitab           = _mm_cvttps_epi32(ewrt);
1169             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1170             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1171                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1172                                          &ewtabF,&ewtabFn);
1173             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1174             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1175
1176             fscal            = felec;
1177
1178             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1179
1180             /* Calculate temporary vectorial force */
1181             tx               = _mm_mul_ps(fscal,dx10);
1182             ty               = _mm_mul_ps(fscal,dy10);
1183             tz               = _mm_mul_ps(fscal,dz10);
1184
1185             /* Update vectorial force */
1186             fix1             = _mm_add_ps(fix1,tx);
1187             fiy1             = _mm_add_ps(fiy1,ty);
1188             fiz1             = _mm_add_ps(fiz1,tz);
1189
1190             fjx0             = _mm_add_ps(fjx0,tx);
1191             fjy0             = _mm_add_ps(fjy0,ty);
1192             fjz0             = _mm_add_ps(fjz0,tz);
1193
1194             /**************************
1195              * CALCULATE INTERACTIONS *
1196              **************************/
1197
1198             r20              = _mm_mul_ps(rsq20,rinv20);
1199             r20              = _mm_andnot_ps(dummy_mask,r20);
1200
1201             /* Compute parameters for interactions between i and j atoms */
1202             qq20             = _mm_mul_ps(iq2,jq0);
1203
1204             /* EWALD ELECTROSTATICS */
1205
1206             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1207             ewrt             = _mm_mul_ps(r20,ewtabscale);
1208             ewitab           = _mm_cvttps_epi32(ewrt);
1209             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1210             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1211                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1212                                          &ewtabF,&ewtabFn);
1213             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1214             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1215
1216             fscal            = felec;
1217
1218             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1219
1220             /* Calculate temporary vectorial force */
1221             tx               = _mm_mul_ps(fscal,dx20);
1222             ty               = _mm_mul_ps(fscal,dy20);
1223             tz               = _mm_mul_ps(fscal,dz20);
1224
1225             /* Update vectorial force */
1226             fix2             = _mm_add_ps(fix2,tx);
1227             fiy2             = _mm_add_ps(fiy2,ty);
1228             fiz2             = _mm_add_ps(fiz2,tz);
1229
1230             fjx0             = _mm_add_ps(fjx0,tx);
1231             fjy0             = _mm_add_ps(fjy0,ty);
1232             fjz0             = _mm_add_ps(fjz0,tz);
1233
1234             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1235             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1236             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1237             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1238
1239             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1240
1241             /* Inner loop uses 134 flops */
1242         }
1243
1244         /* End of innermost loop */
1245
1246         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1247                                               f+i_coord_offset,fshift+i_shift_offset);
1248
1249         /* Increment number of inner iterations */
1250         inneriter                  += j_index_end - j_index_start;
1251
1252         /* Outer loop uses 18 flops */
1253     }
1254
1255     /* Increment number of outer iterations */
1256     outeriter        += nri;
1257
1258     /* Update outer/inner flops */
1259
1260     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*134);
1261 }