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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse2_single.h"
  50 #include "kernelutil_x86_sse2_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3P1_VF_sse2_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_VF_sse2_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 103     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 104     __m128           c6grid_00;
 105     __m128           c6grid_10;
 106     __m128           c6grid_20;
 107     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 108     real             *vdwgridparam;
 109     __m128           one_half = _mm_set1_ps(0.5);
 110     __m128           minus_one = _mm_set1_ps(-1.0);
 111     __m128i          ewitab;
 112     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 113     real             *ewtab;
 114     __m128           dummy_mask,cutoff_mask;
 115     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 116     __m128           one     = _mm_set1_ps(1.0);
 117     __m128           two     = _mm_set1_ps(2.0);
 118     x                = xx[0];
 119     f                = ff[0];
 120
 121     nri              = nlist->nri;
 122     iinr             = nlist->iinr;
 123     jindex           = nlist->jindex;
 124     jjnr             = nlist->jjnr;
 125     shiftidx         = nlist->shift;
 126     gid              = nlist->gid;
 127     shiftvec         = fr->shift_vec[0];
 128     fshift           = fr->fshift[0];
 129     facel            = _mm_set1_ps(fr->epsfac);
 130     charge           = mdatoms->chargeA;
 131     nvdwtype         = fr->ntype;
 132     vdwparam         = fr->nbfp;
 133     vdwtype          = mdatoms->typeA;
 134     vdwgridparam     = fr->ljpme_c6grid;
 135     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 136     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 137     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 138
 139     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 140     ewtab            = fr->ic->tabq_coul_FDV0;
 141     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 142     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 143
 144     /* Setup water-specific parameters */
 145     inr              = nlist->iinr[0];
 146     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 147     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 148     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 149     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 150
 151     /* Avoid stupid compiler warnings */
 152     jnrA = jnrB = jnrC = jnrD = 0;
 153     j_coord_offsetA = 0;
 154     j_coord_offsetB = 0;
 155     j_coord_offsetC = 0;
 156     j_coord_offsetD = 0;
 157
 158     outeriter        = 0;
 159     inneriter        = 0;
 160
 161     for(iidx=0;iidx<4*DIM;iidx++)
 162     {
 163         scratch[iidx] = 0.0;
 164     }
 165
 166     /* Start outer loop over neighborlists */
 167     for(iidx=0; iidx<nri; iidx++)
 168     {
 169         /* Load shift vector for this list */
 170         i_shift_offset   = DIM*shiftidx[iidx];
 171
 172         /* Load limits for loop over neighbors */
 173         j_index_start    = jindex[iidx];
 174         j_index_end      = jindex[iidx+1];
 175
 176         /* Get outer coordinate index */
 177         inr              = iinr[iidx];
 178         i_coord_offset   = DIM*inr;
 179
 180         /* Load i particle coords and add shift vector */
 181         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 182                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 183
 184         fix0             = _mm_setzero_ps();
 185         fiy0             = _mm_setzero_ps();
 186         fiz0             = _mm_setzero_ps();
 187         fix1             = _mm_setzero_ps();
 188         fiy1             = _mm_setzero_ps();
 189         fiz1             = _mm_setzero_ps();
 190         fix2             = _mm_setzero_ps();
 191         fiy2             = _mm_setzero_ps();
 192         fiz2             = _mm_setzero_ps();
 193
 194         /* Reset potential sums */
 195         velecsum         = _mm_setzero_ps();
 196         vvdwsum          = _mm_setzero_ps();
 197
 198         /* Start inner kernel loop */
 199         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 200         {
 201
 202             /* Get j neighbor index, and coordinate index */
 203             jnrA             = jjnr[jidx];
 204             jnrB             = jjnr[jidx+1];
 205             jnrC             = jjnr[jidx+2];
 206             jnrD             = jjnr[jidx+3];
 207             j_coord_offsetA  = DIM*jnrA;
 208             j_coord_offsetB  = DIM*jnrB;
 209             j_coord_offsetC  = DIM*jnrC;
 210             j_coord_offsetD  = DIM*jnrD;
 211
 212             /* load j atom coordinates */
 213             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 214                                               x+j_coord_offsetC,x+j_coord_offsetD,
 215                                               &jx0,&jy0,&jz0);
 216
 217             /* Calculate displacement vector */
 218             dx00             = _mm_sub_ps(ix0,jx0);
 219             dy00             = _mm_sub_ps(iy0,jy0);
 220             dz00             = _mm_sub_ps(iz0,jz0);
 221             dx10             = _mm_sub_ps(ix1,jx0);
 222             dy10             = _mm_sub_ps(iy1,jy0);
 223             dz10             = _mm_sub_ps(iz1,jz0);
 224             dx20             = _mm_sub_ps(ix2,jx0);
 225             dy20             = _mm_sub_ps(iy2,jy0);
 226             dz20             = _mm_sub_ps(iz2,jz0);
 227
 228             /* Calculate squared distance and things based on it */
 229             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 230             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 231             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 232
 233             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 234             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 235             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 236
 237             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 238             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 239             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 240
 241             /* Load parameters for j particles */
 242             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 243                                                               charge+jnrC+0,charge+jnrD+0);
 244             vdwjidx0A        = 2*vdwtype[jnrA+0];
 245             vdwjidx0B        = 2*vdwtype[jnrB+0];
 246             vdwjidx0C        = 2*vdwtype[jnrC+0];
 247             vdwjidx0D        = 2*vdwtype[jnrD+0];
 248
 249             fjx0             = _mm_setzero_ps();
 250             fjy0             = _mm_setzero_ps();
 251             fjz0             = _mm_setzero_ps();
 252
 253             /**************************
 254              * CALCULATE INTERACTIONS *
 255              **************************/
 256
 257             r00              = _mm_mul_ps(rsq00,rinv00);
 258
 259             /* Compute parameters for interactions between i and j atoms */
 260             qq00             = _mm_mul_ps(iq0,jq0);
 261             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 262                                          vdwparam+vdwioffset0+vdwjidx0B,
 263                                          vdwparam+vdwioffset0+vdwjidx0C,
 264                                          vdwparam+vdwioffset0+vdwjidx0D,
 265                                          &c6_00,&c12_00);
 266             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 267                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 268                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 269                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 270
 271             /* EWALD ELECTROSTATICS */
 272
 273             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 274             ewrt             = _mm_mul_ps(r00,ewtabscale);
 275             ewitab           = _mm_cvttps_epi32(ewrt);
 276             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 277             ewitab           = _mm_slli_epi32(ewitab,2);
 278             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 279             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 280             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 281             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 282             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 283             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 284             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 285             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 286             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 287
 288             /* Analytical LJ-PME */
 289             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 290             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 291             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 292             exponent         = gmx_simd_exp_r(ewcljrsq);
 293             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 294             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 295             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 296             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 297             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 298             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 299             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 300             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);
 301
 302             /* Update potential sum for this i atom from the interaction with this j atom. */
 303             velecsum         = _mm_add_ps(velecsum,velec);
 304             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 305
 306             fscal            = _mm_add_ps(felec,fvdw);
 307
 308             /* Calculate temporary vectorial force */
 309             tx               = _mm_mul_ps(fscal,dx00);
 310             ty               = _mm_mul_ps(fscal,dy00);
 311             tz               = _mm_mul_ps(fscal,dz00);
 312
 313             /* Update vectorial force */
 314             fix0             = _mm_add_ps(fix0,tx);
 315             fiy0             = _mm_add_ps(fiy0,ty);
 316             fiz0             = _mm_add_ps(fiz0,tz);
 317
 318             fjx0             = _mm_add_ps(fjx0,tx);
 319             fjy0             = _mm_add_ps(fjy0,ty);
 320             fjz0             = _mm_add_ps(fjz0,tz);
 321
 322             /**************************
 323              * CALCULATE INTERACTIONS *
 324              **************************/
 325
 326             r10              = _mm_mul_ps(rsq10,rinv10);
 327
 328             /* Compute parameters for interactions between i and j atoms */
 329             qq10             = _mm_mul_ps(iq1,jq0);
 330
 331             /* EWALD ELECTROSTATICS */
 332
 333             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 334             ewrt             = _mm_mul_ps(r10,ewtabscale);
 335             ewitab           = _mm_cvttps_epi32(ewrt);
 336             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 337             ewitab           = _mm_slli_epi32(ewitab,2);
 338             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 339             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 340             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 341             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 342             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 343             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 344             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 345             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 346             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 347
 348             /* Update potential sum for this i atom from the interaction with this j atom. */
 349             velecsum         = _mm_add_ps(velecsum,velec);
 350
 351             fscal            = felec;
 352
 353             /* Calculate temporary vectorial force */
 354             tx               = _mm_mul_ps(fscal,dx10);
 355             ty               = _mm_mul_ps(fscal,dy10);
 356             tz               = _mm_mul_ps(fscal,dz10);
 357
 358             /* Update vectorial force */
 359             fix1             = _mm_add_ps(fix1,tx);
 360             fiy1             = _mm_add_ps(fiy1,ty);
 361             fiz1             = _mm_add_ps(fiz1,tz);
 362
 363             fjx0             = _mm_add_ps(fjx0,tx);
 364             fjy0             = _mm_add_ps(fjy0,ty);
 365             fjz0             = _mm_add_ps(fjz0,tz);
 366
 367             /**************************
 368              * CALCULATE INTERACTIONS *
 369              **************************/
 370
 371             r20              = _mm_mul_ps(rsq20,rinv20);
 372
 373             /* Compute parameters for interactions between i and j atoms */
 374             qq20             = _mm_mul_ps(iq2,jq0);
 375
 376             /* EWALD ELECTROSTATICS */
 377
 378             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 379             ewrt             = _mm_mul_ps(r20,ewtabscale);
 380             ewitab           = _mm_cvttps_epi32(ewrt);
 381             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 382             ewitab           = _mm_slli_epi32(ewitab,2);
 383             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 384             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 385             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 386             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 387             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 388             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 389             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 390             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 391             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 392
 393             /* Update potential sum for this i atom from the interaction with this j atom. */
 394             velecsum         = _mm_add_ps(velecsum,velec);
 395
 396             fscal            = felec;
 397
 398             /* Calculate temporary vectorial force */
 399             tx               = _mm_mul_ps(fscal,dx20);
 400             ty               = _mm_mul_ps(fscal,dy20);
 401             tz               = _mm_mul_ps(fscal,dz20);
 402
 403             /* Update vectorial force */
 404             fix2             = _mm_add_ps(fix2,tx);
 405             fiy2             = _mm_add_ps(fiy2,ty);
 406             fiz2             = _mm_add_ps(fiz2,tz);
 407
 408             fjx0             = _mm_add_ps(fjx0,tx);
 409             fjy0             = _mm_add_ps(fjy0,ty);
 410             fjz0             = _mm_add_ps(fjz0,tz);
 411
 412             fjptrA             = f+j_coord_offsetA;
 413             fjptrB             = f+j_coord_offsetB;
 414             fjptrC             = f+j_coord_offsetC;
 415             fjptrD             = f+j_coord_offsetD;
 416
 417             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 418
 419             /* Inner loop uses 151 flops */
 420         }
 421
 422         if(jidx<j_index_end)
 423         {
 424
 425             /* Get j neighbor index, and coordinate index */
 426             jnrlistA         = jjnr[jidx];
 427             jnrlistB         = jjnr[jidx+1];
 428             jnrlistC         = jjnr[jidx+2];
 429             jnrlistD         = jjnr[jidx+3];
 430             /* Sign of each element will be negative for non-real atoms.
 431              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 432              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 433              */
 434             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 435             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 436             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 437             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 438             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 439             j_coord_offsetA  = DIM*jnrA;
 440             j_coord_offsetB  = DIM*jnrB;
 441             j_coord_offsetC  = DIM*jnrC;
 442             j_coord_offsetD  = DIM*jnrD;
 443
 444             /* load j atom coordinates */
 445             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 446                                               x+j_coord_offsetC,x+j_coord_offsetD,
 447                                               &jx0,&jy0,&jz0);
 448
 449             /* Calculate displacement vector */
 450             dx00             = _mm_sub_ps(ix0,jx0);
 451             dy00             = _mm_sub_ps(iy0,jy0);
 452             dz00             = _mm_sub_ps(iz0,jz0);
 453             dx10             = _mm_sub_ps(ix1,jx0);
 454             dy10             = _mm_sub_ps(iy1,jy0);
 455             dz10             = _mm_sub_ps(iz1,jz0);
 456             dx20             = _mm_sub_ps(ix2,jx0);
 457             dy20             = _mm_sub_ps(iy2,jy0);
 458             dz20             = _mm_sub_ps(iz2,jz0);
 459
 460             /* Calculate squared distance and things based on it */
 461             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 462             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 463             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 464
 465             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 466             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 467             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 468
 469             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 470             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 471             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 472
 473             /* Load parameters for j particles */
 474             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 475                                                               charge+jnrC+0,charge+jnrD+0);
 476             vdwjidx0A        = 2*vdwtype[jnrA+0];
 477             vdwjidx0B        = 2*vdwtype[jnrB+0];
 478             vdwjidx0C        = 2*vdwtype[jnrC+0];
 479             vdwjidx0D        = 2*vdwtype[jnrD+0];
 480
 481             fjx0             = _mm_setzero_ps();
 482             fjy0             = _mm_setzero_ps();
 483             fjz0             = _mm_setzero_ps();
 484
 485             /**************************
 486              * CALCULATE INTERACTIONS *
 487              **************************/
 488
 489             r00              = _mm_mul_ps(rsq00,rinv00);
 490             r00              = _mm_andnot_ps(dummy_mask,r00);
 491
 492             /* Compute parameters for interactions between i and j atoms */
 493             qq00             = _mm_mul_ps(iq0,jq0);
 494             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 495                                          vdwparam+vdwioffset0+vdwjidx0B,
 496                                          vdwparam+vdwioffset0+vdwjidx0C,
 497                                          vdwparam+vdwioffset0+vdwjidx0D,
 498                                          &c6_00,&c12_00);
 499             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 500                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 501                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 502                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 503
 504             /* EWALD ELECTROSTATICS */
 505
 506             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 507             ewrt             = _mm_mul_ps(r00,ewtabscale);
 508             ewitab           = _mm_cvttps_epi32(ewrt);
 509             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 510             ewitab           = _mm_slli_epi32(ewitab,2);
 511             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 512             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 513             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 514             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 515             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 516             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 517             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 518             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 519             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 520
 521             /* Analytical LJ-PME */
 522             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 523             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 524             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 525             exponent         = gmx_simd_exp_r(ewcljrsq);
 526             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 527             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 528             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 529             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 530             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 531             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 532             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 533             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);
 534
 535             /* Update potential sum for this i atom from the interaction with this j atom. */
 536             velec            = _mm_andnot_ps(dummy_mask,velec);
 537             velecsum         = _mm_add_ps(velecsum,velec);
 538             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 539             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 540
 541             fscal            = _mm_add_ps(felec,fvdw);
 542
 543             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 544
 545             /* Calculate temporary vectorial force */
 546             tx               = _mm_mul_ps(fscal,dx00);
 547             ty               = _mm_mul_ps(fscal,dy00);
 548             tz               = _mm_mul_ps(fscal,dz00);
 549
 550             /* Update vectorial force */
 551             fix0             = _mm_add_ps(fix0,tx);
 552             fiy0             = _mm_add_ps(fiy0,ty);
 553             fiz0             = _mm_add_ps(fiz0,tz);
 554
 555             fjx0             = _mm_add_ps(fjx0,tx);
 556             fjy0             = _mm_add_ps(fjy0,ty);
 557             fjz0             = _mm_add_ps(fjz0,tz);
 558
 559             /**************************
 560              * CALCULATE INTERACTIONS *
 561              **************************/
 562
 563             r10              = _mm_mul_ps(rsq10,rinv10);
 564             r10              = _mm_andnot_ps(dummy_mask,r10);
 565
 566             /* Compute parameters for interactions between i and j atoms */
 567             qq10             = _mm_mul_ps(iq1,jq0);
 568
 569             /* EWALD ELECTROSTATICS */
 570
 571             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 572             ewrt             = _mm_mul_ps(r10,ewtabscale);
 573             ewitab           = _mm_cvttps_epi32(ewrt);
 574             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 575             ewitab           = _mm_slli_epi32(ewitab,2);
 576             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 577             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 578             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 579             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 580             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 581             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 582             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 583             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 584             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 585
 586             /* Update potential sum for this i atom from the interaction with this j atom. */
 587             velec            = _mm_andnot_ps(dummy_mask,velec);
 588             velecsum         = _mm_add_ps(velecsum,velec);
 589
 590             fscal            = felec;
 591
 592             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 593
 594             /* Calculate temporary vectorial force */
 595             tx               = _mm_mul_ps(fscal,dx10);
 596             ty               = _mm_mul_ps(fscal,dy10);
 597             tz               = _mm_mul_ps(fscal,dz10);
 598
 599             /* Update vectorial force */
 600             fix1             = _mm_add_ps(fix1,tx);
 601             fiy1             = _mm_add_ps(fiy1,ty);
 602             fiz1             = _mm_add_ps(fiz1,tz);
 603
 604             fjx0             = _mm_add_ps(fjx0,tx);
 605             fjy0             = _mm_add_ps(fjy0,ty);
 606             fjz0             = _mm_add_ps(fjz0,tz);
 607
 608             /**************************
 609              * CALCULATE INTERACTIONS *
 610              **************************/
 611
 612             r20              = _mm_mul_ps(rsq20,rinv20);
 613             r20              = _mm_andnot_ps(dummy_mask,r20);
 614
 615             /* Compute parameters for interactions between i and j atoms */
 616             qq20             = _mm_mul_ps(iq2,jq0);
 617
 618             /* EWALD ELECTROSTATICS */
 619
 620             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 621             ewrt             = _mm_mul_ps(r20,ewtabscale);
 622             ewitab           = _mm_cvttps_epi32(ewrt);
 623             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 624             ewitab           = _mm_slli_epi32(ewitab,2);
 625             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 626             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 627             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 628             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 629             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 630             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 631             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 632             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 633             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 634
 635             /* Update potential sum for this i atom from the interaction with this j atom. */
 636             velec            = _mm_andnot_ps(dummy_mask,velec);
 637             velecsum         = _mm_add_ps(velecsum,velec);
 638
 639             fscal            = felec;
 640
 641             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 642
 643             /* Calculate temporary vectorial force */
 644             tx               = _mm_mul_ps(fscal,dx20);
 645             ty               = _mm_mul_ps(fscal,dy20);
 646             tz               = _mm_mul_ps(fscal,dz20);
 647
 648             /* Update vectorial force */
 649             fix2             = _mm_add_ps(fix2,tx);
 650             fiy2             = _mm_add_ps(fiy2,ty);
 651             fiz2             = _mm_add_ps(fiz2,tz);
 652
 653             fjx0             = _mm_add_ps(fjx0,tx);
 654             fjy0             = _mm_add_ps(fjy0,ty);
 655             fjz0             = _mm_add_ps(fjz0,tz);
 656
 657             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 658             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 659             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 660             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 661
 662             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 663
 664             /* Inner loop uses 154 flops */
 665         }
 666
 667         /* End of innermost loop */
 668
 669         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 670                                               f+i_coord_offset,fshift+i_shift_offset);
 671
 672         ggid                        = gid[iidx];
 673         /* Update potential energies */
 674         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 675         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 676
 677         /* Increment number of inner iterations */
 678         inneriter                  += j_index_end - j_index_start;
 679
 680         /* Outer loop uses 20 flops */
 681     }
 682
 683     /* Increment number of outer iterations */
 684     outeriter        += nri;
 685
 686     /* Update outer/inner flops */
 687
 688     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*154);
 689 }
 690 /*
 691  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_sse2_single
 692  * Electrostatics interaction: Ewald
 693  * VdW interaction:            LJEwald
 694  * Geometry:                   Water3-Particle
 695  * Calculate force/pot:        Force
 696  */
 697 void
 698 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_sse2_single
 699                     (t_nblist                    * gmx_restrict       nlist,
 700                      rvec                        * gmx_restrict          xx,
 701                      rvec                        * gmx_restrict          ff,
 702                      t_forcerec                  * gmx_restrict          fr,
 703                      t_mdatoms                   * gmx_restrict     mdatoms,
 704                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 705                      t_nrnb                      * gmx_restrict        nrnb)
 706 {
 707     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 708      * just 0 for non-waters.
 709      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 710      * jnr indices corresponding to data put in the four positions in the SIMD register.
 711      */
 712     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 713     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 714     int              jnrA,jnrB,jnrC,jnrD;
 715     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 716     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 717     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 718     real             rcutoff_scalar;
 719     real             *shiftvec,*fshift,*x,*f;
 720     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 721     real             scratch[4*DIM];
 722     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 723     int              vdwioffset0;
 724     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 725     int              vdwioffset1;
 726     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 727     int              vdwioffset2;
 728     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 729     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 730     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 731     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 732     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 733     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 734     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 735     real             *charge;
 736     int              nvdwtype;
 737     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 738     int              *vdwtype;
 739     real             *vdwparam;
 740     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 741     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 742     __m128           c6grid_00;
 743     __m128           c6grid_10;
 744     __m128           c6grid_20;
 745     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 746     real             *vdwgridparam;
 747     __m128           one_half = _mm_set1_ps(0.5);
 748     __m128           minus_one = _mm_set1_ps(-1.0);
 749     __m128i          ewitab;
 750     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 751     real             *ewtab;
 752     __m128           dummy_mask,cutoff_mask;
 753     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 754     __m128           one     = _mm_set1_ps(1.0);
 755     __m128           two     = _mm_set1_ps(2.0);
 756     x                = xx[0];
 757     f                = ff[0];
 758
 759     nri              = nlist->nri;
 760     iinr             = nlist->iinr;
 761     jindex           = nlist->jindex;
 762     jjnr             = nlist->jjnr;
 763     shiftidx         = nlist->shift;
 764     gid              = nlist->gid;
 765     shiftvec         = fr->shift_vec[0];
 766     fshift           = fr->fshift[0];
 767     facel            = _mm_set1_ps(fr->epsfac);
 768     charge           = mdatoms->chargeA;
 769     nvdwtype         = fr->ntype;
 770     vdwparam         = fr->nbfp;
 771     vdwtype          = mdatoms->typeA;
 772     vdwgridparam     = fr->ljpme_c6grid;
 773     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 774     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 775     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 776
 777     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 778     ewtab            = fr->ic->tabq_coul_F;
 779     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 780     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 781
 782     /* Setup water-specific parameters */
 783     inr              = nlist->iinr[0];
 784     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 785     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 786     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 787     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 788
 789     /* Avoid stupid compiler warnings */
 790     jnrA = jnrB = jnrC = jnrD = 0;
 791     j_coord_offsetA = 0;
 792     j_coord_offsetB = 0;
 793     j_coord_offsetC = 0;
 794     j_coord_offsetD = 0;
 795
 796     outeriter        = 0;
 797     inneriter        = 0;
 798
 799     for(iidx=0;iidx<4*DIM;iidx++)
 800     {
 801         scratch[iidx] = 0.0;
 802     }
 803
 804     /* Start outer loop over neighborlists */
 805     for(iidx=0; iidx<nri; iidx++)
 806     {
 807         /* Load shift vector for this list */
 808         i_shift_offset   = DIM*shiftidx[iidx];
 809
 810         /* Load limits for loop over neighbors */
 811         j_index_start    = jindex[iidx];
 812         j_index_end      = jindex[iidx+1];
 813
 814         /* Get outer coordinate index */
 815         inr              = iinr[iidx];
 816         i_coord_offset   = DIM*inr;
 817
 818         /* Load i particle coords and add shift vector */
 819         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 820                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 821
 822         fix0             = _mm_setzero_ps();
 823         fiy0             = _mm_setzero_ps();
 824         fiz0             = _mm_setzero_ps();
 825         fix1             = _mm_setzero_ps();
 826         fiy1             = _mm_setzero_ps();
 827         fiz1             = _mm_setzero_ps();
 828         fix2             = _mm_setzero_ps();
 829         fiy2             = _mm_setzero_ps();
 830         fiz2             = _mm_setzero_ps();
 831
 832         /* Start inner kernel loop */
 833         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 834         {
 835
 836             /* Get j neighbor index, and coordinate index */
 837             jnrA             = jjnr[jidx];
 838             jnrB             = jjnr[jidx+1];
 839             jnrC             = jjnr[jidx+2];
 840             jnrD             = jjnr[jidx+3];
 841             j_coord_offsetA  = DIM*jnrA;
 842             j_coord_offsetB  = DIM*jnrB;
 843             j_coord_offsetC  = DIM*jnrC;
 844             j_coord_offsetD  = DIM*jnrD;
 845
 846             /* load j atom coordinates */
 847             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 848                                               x+j_coord_offsetC,x+j_coord_offsetD,
 849                                               &jx0,&jy0,&jz0);
 850
 851             /* Calculate displacement vector */
 852             dx00             = _mm_sub_ps(ix0,jx0);
 853             dy00             = _mm_sub_ps(iy0,jy0);
 854             dz00             = _mm_sub_ps(iz0,jz0);
 855             dx10             = _mm_sub_ps(ix1,jx0);
 856             dy10             = _mm_sub_ps(iy1,jy0);
 857             dz10             = _mm_sub_ps(iz1,jz0);
 858             dx20             = _mm_sub_ps(ix2,jx0);
 859             dy20             = _mm_sub_ps(iy2,jy0);
 860             dz20             = _mm_sub_ps(iz2,jz0);
 861
 862             /* Calculate squared distance and things based on it */
 863             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 864             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 865             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 866
 867             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 868             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 869             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 870
 871             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 872             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 873             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 874
 875             /* Load parameters for j particles */
 876             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 877                                                               charge+jnrC+0,charge+jnrD+0);
 878             vdwjidx0A        = 2*vdwtype[jnrA+0];
 879             vdwjidx0B        = 2*vdwtype[jnrB+0];
 880             vdwjidx0C        = 2*vdwtype[jnrC+0];
 881             vdwjidx0D        = 2*vdwtype[jnrD+0];
 882
 883             fjx0             = _mm_setzero_ps();
 884             fjy0             = _mm_setzero_ps();
 885             fjz0             = _mm_setzero_ps();
 886
 887             /**************************
 888              * CALCULATE INTERACTIONS *
 889              **************************/
 890
 891             r00              = _mm_mul_ps(rsq00,rinv00);
 892
 893             /* Compute parameters for interactions between i and j atoms */
 894             qq00             = _mm_mul_ps(iq0,jq0);
 895             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 896                                          vdwparam+vdwioffset0+vdwjidx0B,
 897                                          vdwparam+vdwioffset0+vdwjidx0C,
 898                                          vdwparam+vdwioffset0+vdwjidx0D,
 899                                          &c6_00,&c12_00);
 900             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 901                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 902                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 903                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 904
 905             /* EWALD ELECTROSTATICS */
 906
 907             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 908             ewrt             = _mm_mul_ps(r00,ewtabscale);
 909             ewitab           = _mm_cvttps_epi32(ewrt);
 910             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 911             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 912                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 913                                          &ewtabF,&ewtabFn);
 914             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 915             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 916
 917             /* Analytical LJ-PME */
 918             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 919             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 920             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 921             exponent         = gmx_simd_exp_r(ewcljrsq);
 922             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 923             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 924             /* f6A = 6 * C6grid * (1 - poly) */
 925             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 926             /* f6B = C6grid * exponent * beta^6 */
 927             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 928             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 929             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);
 930
 931             fscal            = _mm_add_ps(felec,fvdw);
 932
 933             /* Calculate temporary vectorial force */
 934             tx               = _mm_mul_ps(fscal,dx00);
 935             ty               = _mm_mul_ps(fscal,dy00);
 936             tz               = _mm_mul_ps(fscal,dz00);
 937
 938             /* Update vectorial force */
 939             fix0             = _mm_add_ps(fix0,tx);
 940             fiy0             = _mm_add_ps(fiy0,ty);
 941             fiz0             = _mm_add_ps(fiz0,tz);
 942
 943             fjx0             = _mm_add_ps(fjx0,tx);
 944             fjy0             = _mm_add_ps(fjy0,ty);
 945             fjz0             = _mm_add_ps(fjz0,tz);
 946
 947             /**************************
 948              * CALCULATE INTERACTIONS *
 949              **************************/
 950
 951             r10              = _mm_mul_ps(rsq10,rinv10);
 952
 953             /* Compute parameters for interactions between i and j atoms */
 954             qq10             = _mm_mul_ps(iq1,jq0);
 955
 956             /* EWALD ELECTROSTATICS */
 957
 958             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 959             ewrt             = _mm_mul_ps(r10,ewtabscale);
 960             ewitab           = _mm_cvttps_epi32(ewrt);
 961             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 962             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 963                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 964                                          &ewtabF,&ewtabFn);
 965             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 966             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 967
 968             fscal            = felec;
 969
 970             /* Calculate temporary vectorial force */
 971             tx               = _mm_mul_ps(fscal,dx10);
 972             ty               = _mm_mul_ps(fscal,dy10);
 973             tz               = _mm_mul_ps(fscal,dz10);
 974
 975             /* Update vectorial force */
 976             fix1             = _mm_add_ps(fix1,tx);
 977             fiy1             = _mm_add_ps(fiy1,ty);
 978             fiz1             = _mm_add_ps(fiz1,tz);
 979
 980             fjx0             = _mm_add_ps(fjx0,tx);
 981             fjy0             = _mm_add_ps(fjy0,ty);
 982             fjz0             = _mm_add_ps(fjz0,tz);
 983
 984             /**************************
 985              * CALCULATE INTERACTIONS *
 986              **************************/
 987
 988             r20              = _mm_mul_ps(rsq20,rinv20);
 989
 990             /* Compute parameters for interactions between i and j atoms */
 991             qq20             = _mm_mul_ps(iq2,jq0);
 992
 993             /* EWALD ELECTROSTATICS */
 994
 995             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 996             ewrt             = _mm_mul_ps(r20,ewtabscale);
 997             ewitab           = _mm_cvttps_epi32(ewrt);
 998             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 999             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1000                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1001                                          &ewtabF,&ewtabFn);
1002             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1003             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1004
1005             fscal            = felec;
1006
1007             /* Calculate temporary vectorial force */
1008             tx               = _mm_mul_ps(fscal,dx20);
1009             ty               = _mm_mul_ps(fscal,dy20);
1010             tz               = _mm_mul_ps(fscal,dz20);
1011
1012             /* Update vectorial force */
1013             fix2             = _mm_add_ps(fix2,tx);
1014             fiy2             = _mm_add_ps(fiy2,ty);
1015             fiz2             = _mm_add_ps(fiz2,tz);
1016
1017             fjx0             = _mm_add_ps(fjx0,tx);
1018             fjy0             = _mm_add_ps(fjy0,ty);
1019             fjz0             = _mm_add_ps(fjz0,tz);
1020
1021             fjptrA             = f+j_coord_offsetA;
1022             fjptrB             = f+j_coord_offsetB;
1023             fjptrC             = f+j_coord_offsetC;
1024             fjptrD             = f+j_coord_offsetD;
1025
1026             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1027
1028             /* Inner loop uses 131 flops */
1029         }
1030
1031         if(jidx<j_index_end)
1032         {
1033
1034             /* Get j neighbor index, and coordinate index */
1035             jnrlistA         = jjnr[jidx];
1036             jnrlistB         = jjnr[jidx+1];
1037             jnrlistC         = jjnr[jidx+2];
1038             jnrlistD         = jjnr[jidx+3];
1039             /* Sign of each element will be negative for non-real atoms.
1040              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1041              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1042              */
1043             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1044             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1045             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1046             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1047             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1048             j_coord_offsetA  = DIM*jnrA;
1049             j_coord_offsetB  = DIM*jnrB;
1050             j_coord_offsetC  = DIM*jnrC;
1051             j_coord_offsetD  = DIM*jnrD;
1052
1053             /* load j atom coordinates */
1054             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1055                                               x+j_coord_offsetC,x+j_coord_offsetD,
1056                                               &jx0,&jy0,&jz0);
1057
1058             /* Calculate displacement vector */
1059             dx00             = _mm_sub_ps(ix0,jx0);
1060             dy00             = _mm_sub_ps(iy0,jy0);
1061             dz00             = _mm_sub_ps(iz0,jz0);
1062             dx10             = _mm_sub_ps(ix1,jx0);
1063             dy10             = _mm_sub_ps(iy1,jy0);
1064             dz10             = _mm_sub_ps(iz1,jz0);
1065             dx20             = _mm_sub_ps(ix2,jx0);
1066             dy20             = _mm_sub_ps(iy2,jy0);
1067             dz20             = _mm_sub_ps(iz2,jz0);
1068
1069             /* Calculate squared distance and things based on it */
1070             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1071             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1072             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1073
1074             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1075             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1076             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1077
1078             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1079             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1080             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1081
1082             /* Load parameters for j particles */
1083             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1084                                                               charge+jnrC+0,charge+jnrD+0);
1085             vdwjidx0A        = 2*vdwtype[jnrA+0];
1086             vdwjidx0B        = 2*vdwtype[jnrB+0];
1087             vdwjidx0C        = 2*vdwtype[jnrC+0];
1088             vdwjidx0D        = 2*vdwtype[jnrD+0];
1089
1090             fjx0             = _mm_setzero_ps();
1091             fjy0             = _mm_setzero_ps();
1092             fjz0             = _mm_setzero_ps();
1093
1094             /**************************
1095              * CALCULATE INTERACTIONS *
1096              **************************/
1097
1098             r00              = _mm_mul_ps(rsq00,rinv00);
1099             r00              = _mm_andnot_ps(dummy_mask,r00);
1100
1101             /* Compute parameters for interactions between i and j atoms */
1102             qq00             = _mm_mul_ps(iq0,jq0);
1103             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1104                                          vdwparam+vdwioffset0+vdwjidx0B,
1105                                          vdwparam+vdwioffset0+vdwjidx0C,
1106                                          vdwparam+vdwioffset0+vdwjidx0D,
1107                                          &c6_00,&c12_00);
1108             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
1109                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
1110                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
1111                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
1112
1113             /* EWALD ELECTROSTATICS */
1114
1115             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1116             ewrt             = _mm_mul_ps(r00,ewtabscale);
1117             ewitab           = _mm_cvttps_epi32(ewrt);
1118             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1119             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1120                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1121                                          &ewtabF,&ewtabFn);
1122             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1123             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1124
1125             /* Analytical LJ-PME */
1126             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1127             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1128             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1129             exponent         = gmx_simd_exp_r(ewcljrsq);
1130             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1131             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1132             /* f6A = 6 * C6grid * (1 - poly) */
1133             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1134             /* f6B = C6grid * exponent * beta^6 */
1135             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1136             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1137             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);
1138
1139             fscal            = _mm_add_ps(felec,fvdw);
1140
1141             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1142
1143             /* Calculate temporary vectorial force */
1144             tx               = _mm_mul_ps(fscal,dx00);
1145             ty               = _mm_mul_ps(fscal,dy00);
1146             tz               = _mm_mul_ps(fscal,dz00);
1147
1148             /* Update vectorial force */
1149             fix0             = _mm_add_ps(fix0,tx);
1150             fiy0             = _mm_add_ps(fiy0,ty);
1151             fiz0             = _mm_add_ps(fiz0,tz);
1152
1153             fjx0             = _mm_add_ps(fjx0,tx);
1154             fjy0             = _mm_add_ps(fjy0,ty);
1155             fjz0             = _mm_add_ps(fjz0,tz);
1156
1157             /**************************
1158              * CALCULATE INTERACTIONS *
1159              **************************/
1160
1161             r10              = _mm_mul_ps(rsq10,rinv10);
1162             r10              = _mm_andnot_ps(dummy_mask,r10);
1163
1164             /* Compute parameters for interactions between i and j atoms */
1165             qq10             = _mm_mul_ps(iq1,jq0);
1166
1167             /* EWALD ELECTROSTATICS */
1168
1169             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1170             ewrt             = _mm_mul_ps(r10,ewtabscale);
1171             ewitab           = _mm_cvttps_epi32(ewrt);
1172             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1173             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1174                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1175                                          &ewtabF,&ewtabFn);
1176             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1177             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1178
1179             fscal            = felec;
1180
1181             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1182
1183             /* Calculate temporary vectorial force */
1184             tx               = _mm_mul_ps(fscal,dx10);
1185             ty               = _mm_mul_ps(fscal,dy10);
1186             tz               = _mm_mul_ps(fscal,dz10);
1187
1188             /* Update vectorial force */
1189             fix1             = _mm_add_ps(fix1,tx);
1190             fiy1             = _mm_add_ps(fiy1,ty);
1191             fiz1             = _mm_add_ps(fiz1,tz);
1192
1193             fjx0             = _mm_add_ps(fjx0,tx);
1194             fjy0             = _mm_add_ps(fjy0,ty);
1195             fjz0             = _mm_add_ps(fjz0,tz);
1196
1197             /**************************
1198              * CALCULATE INTERACTIONS *
1199              **************************/
1200
1201             r20              = _mm_mul_ps(rsq20,rinv20);
1202             r20              = _mm_andnot_ps(dummy_mask,r20);
1203
1204             /* Compute parameters for interactions between i and j atoms */
1205             qq20             = _mm_mul_ps(iq2,jq0);
1206
1207             /* EWALD ELECTROSTATICS */
1208
1209             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1210             ewrt             = _mm_mul_ps(r20,ewtabscale);
1211             ewitab           = _mm_cvttps_epi32(ewrt);
1212             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1213             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1214                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1215                                          &ewtabF,&ewtabFn);
1216             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1217             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1218
1219             fscal            = felec;
1220
1221             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1222
1223             /* Calculate temporary vectorial force */
1224             tx               = _mm_mul_ps(fscal,dx20);
1225             ty               = _mm_mul_ps(fscal,dy20);
1226             tz               = _mm_mul_ps(fscal,dz20);
1227
1228             /* Update vectorial force */
1229             fix2             = _mm_add_ps(fix2,tx);
1230             fiy2             = _mm_add_ps(fiy2,ty);
1231             fiz2             = _mm_add_ps(fiz2,tz);
1232
1233             fjx0             = _mm_add_ps(fjx0,tx);
1234             fjy0             = _mm_add_ps(fjy0,ty);
1235             fjz0             = _mm_add_ps(fjz0,tz);
1236
1237             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1238             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1239             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1240             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1241
1242             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1243
1244             /* Inner loop uses 134 flops */
1245         }
1246
1247         /* End of innermost loop */
1248
1249         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1250                                               f+i_coord_offset,fshift+i_shift_offset);
1251
1252         /* Increment number of inner iterations */
1253         inneriter                  += j_index_end - j_index_start;
1254
1255         /* Outer loop uses 18 flops */
1256     }
1257
1258     /* Increment number of outer iterations */
1259     outeriter        += nri;
1260
1261     /* Update outer/inner flops */
1262
1263     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*134);
1264 }