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, 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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_single.h"
  48 #include "kernelutil_x86_sse2_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3P1_VF_sse2_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LJEwald
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_VF_sse2_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  92     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  93     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  94     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 101     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 102     __m128           c6grid_00;
 103     __m128           c6grid_10;
 104     __m128           c6grid_20;
 105     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 106     real             *vdwgridparam;
 107     __m128           one_half = _mm_set1_ps(0.5);
 108     __m128           minus_one = _mm_set1_ps(-1.0);
 109     __m128i          ewitab;
 110     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 111     real             *ewtab;
 112     __m128           dummy_mask,cutoff_mask;
 113     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 114     __m128           one     = _mm_set1_ps(1.0);
 115     __m128           two     = _mm_set1_ps(2.0);
 116     x                = xx[0];
 117     f                = ff[0];
 118
 119     nri              = nlist->nri;
 120     iinr             = nlist->iinr;
 121     jindex           = nlist->jindex;
 122     jjnr             = nlist->jjnr;
 123     shiftidx         = nlist->shift;
 124     gid              = nlist->gid;
 125     shiftvec         = fr->shift_vec[0];
 126     fshift           = fr->fshift[0];
 127     facel            = _mm_set1_ps(fr->epsfac);
 128     charge           = mdatoms->chargeA;
 129     nvdwtype         = fr->ntype;
 130     vdwparam         = fr->nbfp;
 131     vdwtype          = mdatoms->typeA;
 132     vdwgridparam     = fr->ljpme_c6grid;
 133     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 134     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 135     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 136
 137     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 138     ewtab            = fr->ic->tabq_coul_FDV0;
 139     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 140     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 141
 142     /* Setup water-specific parameters */
 143     inr              = nlist->iinr[0];
 144     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 145     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 146     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 147     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 148
 149     /* Avoid stupid compiler warnings */
 150     jnrA = jnrB = jnrC = jnrD = 0;
 151     j_coord_offsetA = 0;
 152     j_coord_offsetB = 0;
 153     j_coord_offsetC = 0;
 154     j_coord_offsetD = 0;
 155
 156     outeriter        = 0;
 157     inneriter        = 0;
 158
 159     for(iidx=0;iidx<4*DIM;iidx++)
 160     {
 161         scratch[iidx] = 0.0;
 162     }
 163
 164     /* Start outer loop over neighborlists */
 165     for(iidx=0; iidx<nri; iidx++)
 166     {
 167         /* Load shift vector for this list */
 168         i_shift_offset   = DIM*shiftidx[iidx];
 169
 170         /* Load limits for loop over neighbors */
 171         j_index_start    = jindex[iidx];
 172         j_index_end      = jindex[iidx+1];
 173
 174         /* Get outer coordinate index */
 175         inr              = iinr[iidx];
 176         i_coord_offset   = DIM*inr;
 177
 178         /* Load i particle coords and add shift vector */
 179         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 180                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 181
 182         fix0             = _mm_setzero_ps();
 183         fiy0             = _mm_setzero_ps();
 184         fiz0             = _mm_setzero_ps();
 185         fix1             = _mm_setzero_ps();
 186         fiy1             = _mm_setzero_ps();
 187         fiz1             = _mm_setzero_ps();
 188         fix2             = _mm_setzero_ps();
 189         fiy2             = _mm_setzero_ps();
 190         fiz2             = _mm_setzero_ps();
 191
 192         /* Reset potential sums */
 193         velecsum         = _mm_setzero_ps();
 194         vvdwsum          = _mm_setzero_ps();
 195
 196         /* Start inner kernel loop */
 197         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 198         {
 199
 200             /* Get j neighbor index, and coordinate index */
 201             jnrA             = jjnr[jidx];
 202             jnrB             = jjnr[jidx+1];
 203             jnrC             = jjnr[jidx+2];
 204             jnrD             = jjnr[jidx+3];
 205             j_coord_offsetA  = DIM*jnrA;
 206             j_coord_offsetB  = DIM*jnrB;
 207             j_coord_offsetC  = DIM*jnrC;
 208             j_coord_offsetD  = DIM*jnrD;
 209
 210             /* load j atom coordinates */
 211             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 212                                               x+j_coord_offsetC,x+j_coord_offsetD,
 213                                               &jx0,&jy0,&jz0);
 214
 215             /* Calculate displacement vector */
 216             dx00             = _mm_sub_ps(ix0,jx0);
 217             dy00             = _mm_sub_ps(iy0,jy0);
 218             dz00             = _mm_sub_ps(iz0,jz0);
 219             dx10             = _mm_sub_ps(ix1,jx0);
 220             dy10             = _mm_sub_ps(iy1,jy0);
 221             dz10             = _mm_sub_ps(iz1,jz0);
 222             dx20             = _mm_sub_ps(ix2,jx0);
 223             dy20             = _mm_sub_ps(iy2,jy0);
 224             dz20             = _mm_sub_ps(iz2,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
 231             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 232             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 233             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 234
 235             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 236             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 237             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 238
 239             /* Load parameters for j particles */
 240             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 241                                                               charge+jnrC+0,charge+jnrD+0);
 242             vdwjidx0A        = 2*vdwtype[jnrA+0];
 243             vdwjidx0B        = 2*vdwtype[jnrB+0];
 244             vdwjidx0C        = 2*vdwtype[jnrC+0];
 245             vdwjidx0D        = 2*vdwtype[jnrD+0];
 246
 247             fjx0             = _mm_setzero_ps();
 248             fjy0             = _mm_setzero_ps();
 249             fjz0             = _mm_setzero_ps();
 250
 251             /**************************
 252              * CALCULATE INTERACTIONS *
 253              **************************/
 254
 255             r00              = _mm_mul_ps(rsq00,rinv00);
 256
 257             /* Compute parameters for interactions between i and j atoms */
 258             qq00             = _mm_mul_ps(iq0,jq0);
 259             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 260                                          vdwparam+vdwioffset0+vdwjidx0B,
 261                                          vdwparam+vdwioffset0+vdwjidx0C,
 262                                          vdwparam+vdwioffset0+vdwjidx0D,
 263                                          &c6_00,&c12_00);
 264             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 265                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 266                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 267                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 268
 269             /* EWALD ELECTROSTATICS */
 270
 271             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 272             ewrt             = _mm_mul_ps(r00,ewtabscale);
 273             ewitab           = _mm_cvttps_epi32(ewrt);
 274             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 275             ewitab           = _mm_slli_epi32(ewitab,2);
 276             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 277             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 278             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 279             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 280             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 281             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 282             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 283             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 284             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 285
 286             /* Analytical LJ-PME */
 287             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 288             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 289             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 290             exponent         = gmx_simd_exp_r(ewcljrsq);
 291             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 292             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 293             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 294             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 295             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 296             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 297             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 298             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);
 299
 300             /* Update potential sum for this i atom from the interaction with this j atom. */
 301             velecsum         = _mm_add_ps(velecsum,velec);
 302             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 303
 304             fscal            = _mm_add_ps(felec,fvdw);
 305
 306             /* Calculate temporary vectorial force */
 307             tx               = _mm_mul_ps(fscal,dx00);
 308             ty               = _mm_mul_ps(fscal,dy00);
 309             tz               = _mm_mul_ps(fscal,dz00);
 310
 311             /* Update vectorial force */
 312             fix0             = _mm_add_ps(fix0,tx);
 313             fiy0             = _mm_add_ps(fiy0,ty);
 314             fiz0             = _mm_add_ps(fiz0,tz);
 315
 316             fjx0             = _mm_add_ps(fjx0,tx);
 317             fjy0             = _mm_add_ps(fjy0,ty);
 318             fjz0             = _mm_add_ps(fjz0,tz);
 319
 320             /**************************
 321              * CALCULATE INTERACTIONS *
 322              **************************/
 323
 324             r10              = _mm_mul_ps(rsq10,rinv10);
 325
 326             /* Compute parameters for interactions between i and j atoms */
 327             qq10             = _mm_mul_ps(iq1,jq0);
 328
 329             /* EWALD ELECTROSTATICS */
 330
 331             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 332             ewrt             = _mm_mul_ps(r10,ewtabscale);
 333             ewitab           = _mm_cvttps_epi32(ewrt);
 334             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 335             ewitab           = _mm_slli_epi32(ewitab,2);
 336             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 337             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 338             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 339             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 340             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 341             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 342             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 343             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 344             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 345
 346             /* Update potential sum for this i atom from the interaction with this j atom. */
 347             velecsum         = _mm_add_ps(velecsum,velec);
 348
 349             fscal            = felec;
 350
 351             /* Calculate temporary vectorial force */
 352             tx               = _mm_mul_ps(fscal,dx10);
 353             ty               = _mm_mul_ps(fscal,dy10);
 354             tz               = _mm_mul_ps(fscal,dz10);
 355
 356             /* Update vectorial force */
 357             fix1             = _mm_add_ps(fix1,tx);
 358             fiy1             = _mm_add_ps(fiy1,ty);
 359             fiz1             = _mm_add_ps(fiz1,tz);
 360
 361             fjx0             = _mm_add_ps(fjx0,tx);
 362             fjy0             = _mm_add_ps(fjy0,ty);
 363             fjz0             = _mm_add_ps(fjz0,tz);
 364
 365             /**************************
 366              * CALCULATE INTERACTIONS *
 367              **************************/
 368
 369             r20              = _mm_mul_ps(rsq20,rinv20);
 370
 371             /* Compute parameters for interactions between i and j atoms */
 372             qq20             = _mm_mul_ps(iq2,jq0);
 373
 374             /* EWALD ELECTROSTATICS */
 375
 376             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 377             ewrt             = _mm_mul_ps(r20,ewtabscale);
 378             ewitab           = _mm_cvttps_epi32(ewrt);
 379             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 380             ewitab           = _mm_slli_epi32(ewitab,2);
 381             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 382             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 383             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 384             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 385             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 386             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 387             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 388             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 389             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 390
 391             /* Update potential sum for this i atom from the interaction with this j atom. */
 392             velecsum         = _mm_add_ps(velecsum,velec);
 393
 394             fscal            = felec;
 395
 396             /* Calculate temporary vectorial force */
 397             tx               = _mm_mul_ps(fscal,dx20);
 398             ty               = _mm_mul_ps(fscal,dy20);
 399             tz               = _mm_mul_ps(fscal,dz20);
 400
 401             /* Update vectorial force */
 402             fix2             = _mm_add_ps(fix2,tx);
 403             fiy2             = _mm_add_ps(fiy2,ty);
 404             fiz2             = _mm_add_ps(fiz2,tz);
 405
 406             fjx0             = _mm_add_ps(fjx0,tx);
 407             fjy0             = _mm_add_ps(fjy0,ty);
 408             fjz0             = _mm_add_ps(fjz0,tz);
 409
 410             fjptrA             = f+j_coord_offsetA;
 411             fjptrB             = f+j_coord_offsetB;
 412             fjptrC             = f+j_coord_offsetC;
 413             fjptrD             = f+j_coord_offsetD;
 414
 415             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 416
 417             /* Inner loop uses 151 flops */
 418         }
 419
 420         if(jidx<j_index_end)
 421         {
 422
 423             /* Get j neighbor index, and coordinate index */
 424             jnrlistA         = jjnr[jidx];
 425             jnrlistB         = jjnr[jidx+1];
 426             jnrlistC         = jjnr[jidx+2];
 427             jnrlistD         = jjnr[jidx+3];
 428             /* Sign of each element will be negative for non-real atoms.
 429              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 430              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 431              */
 432             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 433             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 434             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 435             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 436             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 437             j_coord_offsetA  = DIM*jnrA;
 438             j_coord_offsetB  = DIM*jnrB;
 439             j_coord_offsetC  = DIM*jnrC;
 440             j_coord_offsetD  = DIM*jnrD;
 441
 442             /* load j atom coordinates */
 443             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 444                                               x+j_coord_offsetC,x+j_coord_offsetD,
 445                                               &jx0,&jy0,&jz0);
 446
 447             /* Calculate displacement vector */
 448             dx00             = _mm_sub_ps(ix0,jx0);
 449             dy00             = _mm_sub_ps(iy0,jy0);
 450             dz00             = _mm_sub_ps(iz0,jz0);
 451             dx10             = _mm_sub_ps(ix1,jx0);
 452             dy10             = _mm_sub_ps(iy1,jy0);
 453             dz10             = _mm_sub_ps(iz1,jz0);
 454             dx20             = _mm_sub_ps(ix2,jx0);
 455             dy20             = _mm_sub_ps(iy2,jy0);
 456             dz20             = _mm_sub_ps(iz2,jz0);
 457
 458             /* Calculate squared distance and things based on it */
 459             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 460             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 461             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 462
 463             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 464             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 465             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 466
 467             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 468             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 469             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 470
 471             /* Load parameters for j particles */
 472             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 473                                                               charge+jnrC+0,charge+jnrD+0);
 474             vdwjidx0A        = 2*vdwtype[jnrA+0];
 475             vdwjidx0B        = 2*vdwtype[jnrB+0];
 476             vdwjidx0C        = 2*vdwtype[jnrC+0];
 477             vdwjidx0D        = 2*vdwtype[jnrD+0];
 478
 479             fjx0             = _mm_setzero_ps();
 480             fjy0             = _mm_setzero_ps();
 481             fjz0             = _mm_setzero_ps();
 482
 483             /**************************
 484              * CALCULATE INTERACTIONS *
 485              **************************/
 486
 487             r00              = _mm_mul_ps(rsq00,rinv00);
 488             r00              = _mm_andnot_ps(dummy_mask,r00);
 489
 490             /* Compute parameters for interactions between i and j atoms */
 491             qq00             = _mm_mul_ps(iq0,jq0);
 492             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 493                                          vdwparam+vdwioffset0+vdwjidx0B,
 494                                          vdwparam+vdwioffset0+vdwjidx0C,
 495                                          vdwparam+vdwioffset0+vdwjidx0D,
 496                                          &c6_00,&c12_00);
 497             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 498                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 499                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 500                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 501
 502             /* EWALD ELECTROSTATICS */
 503
 504             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 505             ewrt             = _mm_mul_ps(r00,ewtabscale);
 506             ewitab           = _mm_cvttps_epi32(ewrt);
 507             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 508             ewitab           = _mm_slli_epi32(ewitab,2);
 509             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 510             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 511             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 512             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 513             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 514             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 515             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 516             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 517             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 518
 519             /* Analytical LJ-PME */
 520             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 521             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 522             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 523             exponent         = gmx_simd_exp_r(ewcljrsq);
 524             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 525             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 526             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 527             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 528             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 529             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 530             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 531             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);
 532
 533             /* Update potential sum for this i atom from the interaction with this j atom. */
 534             velec            = _mm_andnot_ps(dummy_mask,velec);
 535             velecsum         = _mm_add_ps(velecsum,velec);
 536             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 537             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 538
 539             fscal            = _mm_add_ps(felec,fvdw);
 540
 541             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 542
 543             /* Calculate temporary vectorial force */
 544             tx               = _mm_mul_ps(fscal,dx00);
 545             ty               = _mm_mul_ps(fscal,dy00);
 546             tz               = _mm_mul_ps(fscal,dz00);
 547
 548             /* Update vectorial force */
 549             fix0             = _mm_add_ps(fix0,tx);
 550             fiy0             = _mm_add_ps(fiy0,ty);
 551             fiz0             = _mm_add_ps(fiz0,tz);
 552
 553             fjx0             = _mm_add_ps(fjx0,tx);
 554             fjy0             = _mm_add_ps(fjy0,ty);
 555             fjz0             = _mm_add_ps(fjz0,tz);
 556
 557             /**************************
 558              * CALCULATE INTERACTIONS *
 559              **************************/
 560
 561             r10              = _mm_mul_ps(rsq10,rinv10);
 562             r10              = _mm_andnot_ps(dummy_mask,r10);
 563
 564             /* Compute parameters for interactions between i and j atoms */
 565             qq10             = _mm_mul_ps(iq1,jq0);
 566
 567             /* EWALD ELECTROSTATICS */
 568
 569             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 570             ewrt             = _mm_mul_ps(r10,ewtabscale);
 571             ewitab           = _mm_cvttps_epi32(ewrt);
 572             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 573             ewitab           = _mm_slli_epi32(ewitab,2);
 574             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 575             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 576             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 577             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 578             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 579             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 580             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 581             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 582             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 583
 584             /* Update potential sum for this i atom from the interaction with this j atom. */
 585             velec            = _mm_andnot_ps(dummy_mask,velec);
 586             velecsum         = _mm_add_ps(velecsum,velec);
 587
 588             fscal            = felec;
 589
 590             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 591
 592             /* Calculate temporary vectorial force */
 593             tx               = _mm_mul_ps(fscal,dx10);
 594             ty               = _mm_mul_ps(fscal,dy10);
 595             tz               = _mm_mul_ps(fscal,dz10);
 596
 597             /* Update vectorial force */
 598             fix1             = _mm_add_ps(fix1,tx);
 599             fiy1             = _mm_add_ps(fiy1,ty);
 600             fiz1             = _mm_add_ps(fiz1,tz);
 601
 602             fjx0             = _mm_add_ps(fjx0,tx);
 603             fjy0             = _mm_add_ps(fjy0,ty);
 604             fjz0             = _mm_add_ps(fjz0,tz);
 605
 606             /**************************
 607              * CALCULATE INTERACTIONS *
 608              **************************/
 609
 610             r20              = _mm_mul_ps(rsq20,rinv20);
 611             r20              = _mm_andnot_ps(dummy_mask,r20);
 612
 613             /* Compute parameters for interactions between i and j atoms */
 614             qq20             = _mm_mul_ps(iq2,jq0);
 615
 616             /* EWALD ELECTROSTATICS */
 617
 618             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 619             ewrt             = _mm_mul_ps(r20,ewtabscale);
 620             ewitab           = _mm_cvttps_epi32(ewrt);
 621             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 622             ewitab           = _mm_slli_epi32(ewitab,2);
 623             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 624             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 625             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 626             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 627             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 628             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 629             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 630             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 631             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 632
 633             /* Update potential sum for this i atom from the interaction with this j atom. */
 634             velec            = _mm_andnot_ps(dummy_mask,velec);
 635             velecsum         = _mm_add_ps(velecsum,velec);
 636
 637             fscal            = felec;
 638
 639             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 640
 641             /* Calculate temporary vectorial force */
 642             tx               = _mm_mul_ps(fscal,dx20);
 643             ty               = _mm_mul_ps(fscal,dy20);
 644             tz               = _mm_mul_ps(fscal,dz20);
 645
 646             /* Update vectorial force */
 647             fix2             = _mm_add_ps(fix2,tx);
 648             fiy2             = _mm_add_ps(fiy2,ty);
 649             fiz2             = _mm_add_ps(fiz2,tz);
 650
 651             fjx0             = _mm_add_ps(fjx0,tx);
 652             fjy0             = _mm_add_ps(fjy0,ty);
 653             fjz0             = _mm_add_ps(fjz0,tz);
 654
 655             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 656             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 657             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 658             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 659
 660             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 661
 662             /* Inner loop uses 154 flops */
 663         }
 664
 665         /* End of innermost loop */
 666
 667         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 668                                               f+i_coord_offset,fshift+i_shift_offset);
 669
 670         ggid                        = gid[iidx];
 671         /* Update potential energies */
 672         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 673         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 674
 675         /* Increment number of inner iterations */
 676         inneriter                  += j_index_end - j_index_start;
 677
 678         /* Outer loop uses 20 flops */
 679     }
 680
 681     /* Increment number of outer iterations */
 682     outeriter        += nri;
 683
 684     /* Update outer/inner flops */
 685
 686     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*154);
 687 }
 688 /*
 689  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_sse2_single
 690  * Electrostatics interaction: Ewald
 691  * VdW interaction:            LJEwald
 692  * Geometry:                   Water3-Particle
 693  * Calculate force/pot:        Force
 694  */
 695 void
 696 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_sse2_single
 697                     (t_nblist                    * gmx_restrict       nlist,
 698                      rvec                        * gmx_restrict          xx,
 699                      rvec                        * gmx_restrict          ff,
 700                      t_forcerec                  * gmx_restrict          fr,
 701                      t_mdatoms                   * gmx_restrict     mdatoms,
 702                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 703                      t_nrnb                      * gmx_restrict        nrnb)
 704 {
 705     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 706      * just 0 for non-waters.
 707      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 708      * jnr indices corresponding to data put in the four positions in the SIMD register.
 709      */
 710     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 711     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 712     int              jnrA,jnrB,jnrC,jnrD;
 713     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 714     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 715     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 716     real             rcutoff_scalar;
 717     real             *shiftvec,*fshift,*x,*f;
 718     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 719     real             scratch[4*DIM];
 720     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 721     int              vdwioffset0;
 722     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 723     int              vdwioffset1;
 724     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 725     int              vdwioffset2;
 726     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 727     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 728     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 729     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 730     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 731     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 732     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 733     real             *charge;
 734     int              nvdwtype;
 735     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 736     int              *vdwtype;
 737     real             *vdwparam;
 738     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 739     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 740     __m128           c6grid_00;
 741     __m128           c6grid_10;
 742     __m128           c6grid_20;
 743     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 744     real             *vdwgridparam;
 745     __m128           one_half = _mm_set1_ps(0.5);
 746     __m128           minus_one = _mm_set1_ps(-1.0);
 747     __m128i          ewitab;
 748     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 749     real             *ewtab;
 750     __m128           dummy_mask,cutoff_mask;
 751     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 752     __m128           one     = _mm_set1_ps(1.0);
 753     __m128           two     = _mm_set1_ps(2.0);
 754     x                = xx[0];
 755     f                = ff[0];
 756
 757     nri              = nlist->nri;
 758     iinr             = nlist->iinr;
 759     jindex           = nlist->jindex;
 760     jjnr             = nlist->jjnr;
 761     shiftidx         = nlist->shift;
 762     gid              = nlist->gid;
 763     shiftvec         = fr->shift_vec[0];
 764     fshift           = fr->fshift[0];
 765     facel            = _mm_set1_ps(fr->epsfac);
 766     charge           = mdatoms->chargeA;
 767     nvdwtype         = fr->ntype;
 768     vdwparam         = fr->nbfp;
 769     vdwtype          = mdatoms->typeA;
 770     vdwgridparam     = fr->ljpme_c6grid;
 771     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 772     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 773     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 774
 775     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 776     ewtab            = fr->ic->tabq_coul_F;
 777     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 778     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 779
 780     /* Setup water-specific parameters */
 781     inr              = nlist->iinr[0];
 782     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 783     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 784     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 785     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 786
 787     /* Avoid stupid compiler warnings */
 788     jnrA = jnrB = jnrC = jnrD = 0;
 789     j_coord_offsetA = 0;
 790     j_coord_offsetB = 0;
 791     j_coord_offsetC = 0;
 792     j_coord_offsetD = 0;
 793
 794     outeriter        = 0;
 795     inneriter        = 0;
 796
 797     for(iidx=0;iidx<4*DIM;iidx++)
 798     {
 799         scratch[iidx] = 0.0;
 800     }
 801
 802     /* Start outer loop over neighborlists */
 803     for(iidx=0; iidx<nri; iidx++)
 804     {
 805         /* Load shift vector for this list */
 806         i_shift_offset   = DIM*shiftidx[iidx];
 807
 808         /* Load limits for loop over neighbors */
 809         j_index_start    = jindex[iidx];
 810         j_index_end      = jindex[iidx+1];
 811
 812         /* Get outer coordinate index */
 813         inr              = iinr[iidx];
 814         i_coord_offset   = DIM*inr;
 815
 816         /* Load i particle coords and add shift vector */
 817         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 818                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 819
 820         fix0             = _mm_setzero_ps();
 821         fiy0             = _mm_setzero_ps();
 822         fiz0             = _mm_setzero_ps();
 823         fix1             = _mm_setzero_ps();
 824         fiy1             = _mm_setzero_ps();
 825         fiz1             = _mm_setzero_ps();
 826         fix2             = _mm_setzero_ps();
 827         fiy2             = _mm_setzero_ps();
 828         fiz2             = _mm_setzero_ps();
 829
 830         /* Start inner kernel loop */
 831         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 832         {
 833
 834             /* Get j neighbor index, and coordinate index */
 835             jnrA             = jjnr[jidx];
 836             jnrB             = jjnr[jidx+1];
 837             jnrC             = jjnr[jidx+2];
 838             jnrD             = jjnr[jidx+3];
 839             j_coord_offsetA  = DIM*jnrA;
 840             j_coord_offsetB  = DIM*jnrB;
 841             j_coord_offsetC  = DIM*jnrC;
 842             j_coord_offsetD  = DIM*jnrD;
 843
 844             /* load j atom coordinates */
 845             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 846                                               x+j_coord_offsetC,x+j_coord_offsetD,
 847                                               &jx0,&jy0,&jz0);
 848
 849             /* Calculate displacement vector */
 850             dx00             = _mm_sub_ps(ix0,jx0);
 851             dy00             = _mm_sub_ps(iy0,jy0);
 852             dz00             = _mm_sub_ps(iz0,jz0);
 853             dx10             = _mm_sub_ps(ix1,jx0);
 854             dy10             = _mm_sub_ps(iy1,jy0);
 855             dz10             = _mm_sub_ps(iz1,jz0);
 856             dx20             = _mm_sub_ps(ix2,jx0);
 857             dy20             = _mm_sub_ps(iy2,jy0);
 858             dz20             = _mm_sub_ps(iz2,jz0);
 859
 860             /* Calculate squared distance and things based on it */
 861             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 862             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 863             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 864
 865             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 866             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 867             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 868
 869             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 870             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 871             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 872
 873             /* Load parameters for j particles */
 874             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 875                                                               charge+jnrC+0,charge+jnrD+0);
 876             vdwjidx0A        = 2*vdwtype[jnrA+0];
 877             vdwjidx0B        = 2*vdwtype[jnrB+0];
 878             vdwjidx0C        = 2*vdwtype[jnrC+0];
 879             vdwjidx0D        = 2*vdwtype[jnrD+0];
 880
 881             fjx0             = _mm_setzero_ps();
 882             fjy0             = _mm_setzero_ps();
 883             fjz0             = _mm_setzero_ps();
 884
 885             /**************************
 886              * CALCULATE INTERACTIONS *
 887              **************************/
 888
 889             r00              = _mm_mul_ps(rsq00,rinv00);
 890
 891             /* Compute parameters for interactions between i and j atoms */
 892             qq00             = _mm_mul_ps(iq0,jq0);
 893             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 894                                          vdwparam+vdwioffset0+vdwjidx0B,
 895                                          vdwparam+vdwioffset0+vdwjidx0C,
 896                                          vdwparam+vdwioffset0+vdwjidx0D,
 897                                          &c6_00,&c12_00);
 898             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 899                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 900                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 901                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 902
 903             /* EWALD ELECTROSTATICS */
 904
 905             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 906             ewrt             = _mm_mul_ps(r00,ewtabscale);
 907             ewitab           = _mm_cvttps_epi32(ewrt);
 908             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 909             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 910                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 911                                          &ewtabF,&ewtabFn);
 912             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 913             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 914
 915             /* Analytical LJ-PME */
 916             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 917             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 918             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 919             exponent         = gmx_simd_exp_r(ewcljrsq);
 920             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 921             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 922             /* f6A = 6 * C6grid * (1 - poly) */
 923             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 924             /* f6B = C6grid * exponent * beta^6 */
 925             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 926             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 927             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);
 928
 929             fscal            = _mm_add_ps(felec,fvdw);
 930
 931             /* Calculate temporary vectorial force */
 932             tx               = _mm_mul_ps(fscal,dx00);
 933             ty               = _mm_mul_ps(fscal,dy00);
 934             tz               = _mm_mul_ps(fscal,dz00);
 935
 936             /* Update vectorial force */
 937             fix0             = _mm_add_ps(fix0,tx);
 938             fiy0             = _mm_add_ps(fiy0,ty);
 939             fiz0             = _mm_add_ps(fiz0,tz);
 940
 941             fjx0             = _mm_add_ps(fjx0,tx);
 942             fjy0             = _mm_add_ps(fjy0,ty);
 943             fjz0             = _mm_add_ps(fjz0,tz);
 944
 945             /**************************
 946              * CALCULATE INTERACTIONS *
 947              **************************/
 948
 949             r10              = _mm_mul_ps(rsq10,rinv10);
 950
 951             /* Compute parameters for interactions between i and j atoms */
 952             qq10             = _mm_mul_ps(iq1,jq0);
 953
 954             /* EWALD ELECTROSTATICS */
 955
 956             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 957             ewrt             = _mm_mul_ps(r10,ewtabscale);
 958             ewitab           = _mm_cvttps_epi32(ewrt);
 959             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 960             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 961                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 962                                          &ewtabF,&ewtabFn);
 963             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 964             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 965
 966             fscal            = felec;
 967
 968             /* Calculate temporary vectorial force */
 969             tx               = _mm_mul_ps(fscal,dx10);
 970             ty               = _mm_mul_ps(fscal,dy10);
 971             tz               = _mm_mul_ps(fscal,dz10);
 972
 973             /* Update vectorial force */
 974             fix1             = _mm_add_ps(fix1,tx);
 975             fiy1             = _mm_add_ps(fiy1,ty);
 976             fiz1             = _mm_add_ps(fiz1,tz);
 977
 978             fjx0             = _mm_add_ps(fjx0,tx);
 979             fjy0             = _mm_add_ps(fjy0,ty);
 980             fjz0             = _mm_add_ps(fjz0,tz);
 981
 982             /**************************
 983              * CALCULATE INTERACTIONS *
 984              **************************/
 985
 986             r20              = _mm_mul_ps(rsq20,rinv20);
 987
 988             /* Compute parameters for interactions between i and j atoms */
 989             qq20             = _mm_mul_ps(iq2,jq0);
 990
 991             /* EWALD ELECTROSTATICS */
 992
 993             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 994             ewrt             = _mm_mul_ps(r20,ewtabscale);
 995             ewitab           = _mm_cvttps_epi32(ewrt);
 996             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 997             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 998                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 999                                          &ewtabF,&ewtabFn);
1000             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1001             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1002
1003             fscal            = felec;
1004
1005             /* Calculate temporary vectorial force */
1006             tx               = _mm_mul_ps(fscal,dx20);
1007             ty               = _mm_mul_ps(fscal,dy20);
1008             tz               = _mm_mul_ps(fscal,dz20);
1009
1010             /* Update vectorial force */
1011             fix2             = _mm_add_ps(fix2,tx);
1012             fiy2             = _mm_add_ps(fiy2,ty);
1013             fiz2             = _mm_add_ps(fiz2,tz);
1014
1015             fjx0             = _mm_add_ps(fjx0,tx);
1016             fjy0             = _mm_add_ps(fjy0,ty);
1017             fjz0             = _mm_add_ps(fjz0,tz);
1018
1019             fjptrA             = f+j_coord_offsetA;
1020             fjptrB             = f+j_coord_offsetB;
1021             fjptrC             = f+j_coord_offsetC;
1022             fjptrD             = f+j_coord_offsetD;
1023
1024             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1025
1026             /* Inner loop uses 131 flops */
1027         }
1028
1029         if(jidx<j_index_end)
1030         {
1031
1032             /* Get j neighbor index, and coordinate index */
1033             jnrlistA         = jjnr[jidx];
1034             jnrlistB         = jjnr[jidx+1];
1035             jnrlistC         = jjnr[jidx+2];
1036             jnrlistD         = jjnr[jidx+3];
1037             /* Sign of each element will be negative for non-real atoms.
1038              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1039              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1040              */
1041             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1042             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1043             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1044             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1045             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1046             j_coord_offsetA  = DIM*jnrA;
1047             j_coord_offsetB  = DIM*jnrB;
1048             j_coord_offsetC  = DIM*jnrC;
1049             j_coord_offsetD  = DIM*jnrD;
1050
1051             /* load j atom coordinates */
1052             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1053                                               x+j_coord_offsetC,x+j_coord_offsetD,
1054                                               &jx0,&jy0,&jz0);
1055
1056             /* Calculate displacement vector */
1057             dx00             = _mm_sub_ps(ix0,jx0);
1058             dy00             = _mm_sub_ps(iy0,jy0);
1059             dz00             = _mm_sub_ps(iz0,jz0);
1060             dx10             = _mm_sub_ps(ix1,jx0);
1061             dy10             = _mm_sub_ps(iy1,jy0);
1062             dz10             = _mm_sub_ps(iz1,jz0);
1063             dx20             = _mm_sub_ps(ix2,jx0);
1064             dy20             = _mm_sub_ps(iy2,jy0);
1065             dz20             = _mm_sub_ps(iz2,jz0);
1066
1067             /* Calculate squared distance and things based on it */
1068             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1069             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1070             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1071
1072             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1073             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1074             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1075
1076             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1077             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1078             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1079
1080             /* Load parameters for j particles */
1081             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1082                                                               charge+jnrC+0,charge+jnrD+0);
1083             vdwjidx0A        = 2*vdwtype[jnrA+0];
1084             vdwjidx0B        = 2*vdwtype[jnrB+0];
1085             vdwjidx0C        = 2*vdwtype[jnrC+0];
1086             vdwjidx0D        = 2*vdwtype[jnrD+0];
1087
1088             fjx0             = _mm_setzero_ps();
1089             fjy0             = _mm_setzero_ps();
1090             fjz0             = _mm_setzero_ps();
1091
1092             /**************************
1093              * CALCULATE INTERACTIONS *
1094              **************************/
1095
1096             r00              = _mm_mul_ps(rsq00,rinv00);
1097             r00              = _mm_andnot_ps(dummy_mask,r00);
1098
1099             /* Compute parameters for interactions between i and j atoms */
1100             qq00             = _mm_mul_ps(iq0,jq0);
1101             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1102                                          vdwparam+vdwioffset0+vdwjidx0B,
1103                                          vdwparam+vdwioffset0+vdwjidx0C,
1104                                          vdwparam+vdwioffset0+vdwjidx0D,
1105                                          &c6_00,&c12_00);
1106             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
1107                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
1108                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
1109                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
1110
1111             /* EWALD ELECTROSTATICS */
1112
1113             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1114             ewrt             = _mm_mul_ps(r00,ewtabscale);
1115             ewitab           = _mm_cvttps_epi32(ewrt);
1116             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1117             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1118                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1119                                          &ewtabF,&ewtabFn);
1120             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1121             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1122
1123             /* Analytical LJ-PME */
1124             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1125             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1126             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1127             exponent         = gmx_simd_exp_r(ewcljrsq);
1128             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1129             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1130             /* f6A = 6 * C6grid * (1 - poly) */
1131             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1132             /* f6B = C6grid * exponent * beta^6 */
1133             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1134             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1135             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);
1136
1137             fscal            = _mm_add_ps(felec,fvdw);
1138
1139             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1140
1141             /* Calculate temporary vectorial force */
1142             tx               = _mm_mul_ps(fscal,dx00);
1143             ty               = _mm_mul_ps(fscal,dy00);
1144             tz               = _mm_mul_ps(fscal,dz00);
1145
1146             /* Update vectorial force */
1147             fix0             = _mm_add_ps(fix0,tx);
1148             fiy0             = _mm_add_ps(fiy0,ty);
1149             fiz0             = _mm_add_ps(fiz0,tz);
1150
1151             fjx0             = _mm_add_ps(fjx0,tx);
1152             fjy0             = _mm_add_ps(fjy0,ty);
1153             fjz0             = _mm_add_ps(fjz0,tz);
1154
1155             /**************************
1156              * CALCULATE INTERACTIONS *
1157              **************************/
1158
1159             r10              = _mm_mul_ps(rsq10,rinv10);
1160             r10              = _mm_andnot_ps(dummy_mask,r10);
1161
1162             /* Compute parameters for interactions between i and j atoms */
1163             qq10             = _mm_mul_ps(iq1,jq0);
1164
1165             /* EWALD ELECTROSTATICS */
1166
1167             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1168             ewrt             = _mm_mul_ps(r10,ewtabscale);
1169             ewitab           = _mm_cvttps_epi32(ewrt);
1170             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1171             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1172                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1173                                          &ewtabF,&ewtabFn);
1174             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1175             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1176
1177             fscal            = felec;
1178
1179             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1180
1181             /* Calculate temporary vectorial force */
1182             tx               = _mm_mul_ps(fscal,dx10);
1183             ty               = _mm_mul_ps(fscal,dy10);
1184             tz               = _mm_mul_ps(fscal,dz10);
1185
1186             /* Update vectorial force */
1187             fix1             = _mm_add_ps(fix1,tx);
1188             fiy1             = _mm_add_ps(fiy1,ty);
1189             fiz1             = _mm_add_ps(fiz1,tz);
1190
1191             fjx0             = _mm_add_ps(fjx0,tx);
1192             fjy0             = _mm_add_ps(fjy0,ty);
1193             fjz0             = _mm_add_ps(fjz0,tz);
1194
1195             /**************************
1196              * CALCULATE INTERACTIONS *
1197              **************************/
1198
1199             r20              = _mm_mul_ps(rsq20,rinv20);
1200             r20              = _mm_andnot_ps(dummy_mask,r20);
1201
1202             /* Compute parameters for interactions between i and j atoms */
1203             qq20             = _mm_mul_ps(iq2,jq0);
1204
1205             /* EWALD ELECTROSTATICS */
1206
1207             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1208             ewrt             = _mm_mul_ps(r20,ewtabscale);
1209             ewitab           = _mm_cvttps_epi32(ewrt);
1210             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1211             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1212                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1213                                          &ewtabF,&ewtabFn);
1214             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1215             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1216
1217             fscal            = felec;
1218
1219             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1220
1221             /* Calculate temporary vectorial force */
1222             tx               = _mm_mul_ps(fscal,dx20);
1223             ty               = _mm_mul_ps(fscal,dy20);
1224             tz               = _mm_mul_ps(fscal,dz20);
1225
1226             /* Update vectorial force */
1227             fix2             = _mm_add_ps(fix2,tx);
1228             fiy2             = _mm_add_ps(fiy2,ty);
1229             fiz2             = _mm_add_ps(fiz2,tz);
1230
1231             fjx0             = _mm_add_ps(fjx0,tx);
1232             fjy0             = _mm_add_ps(fjy0,ty);
1233             fjz0             = _mm_add_ps(fjz0,tz);
1234
1235             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1236             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1237             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1238             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1239
1240             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1241
1242             /* Inner loop uses 134 flops */
1243         }
1244
1245         /* End of innermost loop */
1246
1247         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1248                                               f+i_coord_offset,fshift+i_shift_offset);
1249
1250         /* Increment number of inner iterations */
1251         inneriter                  += j_index_end - j_index_start;
1252
1253         /* Outer loop uses 18 flops */
1254     }
1255
1256     /* Increment number of outer iterations */
1257     outeriter        += nri;
1258
1259     /* Update outer/inner flops */
1260
1261     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*134);
1262 }