src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEw_VdwLJEw_GeomW3P1_sse4_1_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 sse4_1_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse4_1_single.h"
  50 #include "kernelutil_x86_sse4_1_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3P1_VF_sse4_1_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_sse4_1_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              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
 267             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 268                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 269                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 270                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 271
 272             /* EWALD ELECTROSTATICS */
 273
 274             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 275             ewrt             = _mm_mul_ps(r00,ewtabscale);
 276             ewitab           = _mm_cvttps_epi32(ewrt);
 277             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 278             ewitab           = _mm_slli_epi32(ewitab,2);
 279             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 280             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 281             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 282             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 283             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 284             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 285             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 286             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 287             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 288
 289             /* Analytical LJ-PME */
 290             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 291             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 292             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 293             exponent         = gmx_simd_exp_r(ewcljrsq);
 294             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 295             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 296             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 297             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 298             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 299             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 300             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 301             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);
 302
 303             /* Update potential sum for this i atom from the interaction with this j atom. */
 304             velecsum         = _mm_add_ps(velecsum,velec);
 305             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 306
 307             fscal            = _mm_add_ps(felec,fvdw);
 308
 309             /* Calculate temporary vectorial force */
 310             tx               = _mm_mul_ps(fscal,dx00);
 311             ty               = _mm_mul_ps(fscal,dy00);
 312             tz               = _mm_mul_ps(fscal,dz00);
 313
 314             /* Update vectorial force */
 315             fix0             = _mm_add_ps(fix0,tx);
 316             fiy0             = _mm_add_ps(fiy0,ty);
 317             fiz0             = _mm_add_ps(fiz0,tz);
 318
 319             fjx0             = _mm_add_ps(fjx0,tx);
 320             fjy0             = _mm_add_ps(fjy0,ty);
 321             fjz0             = _mm_add_ps(fjz0,tz);
 322
 323             /**************************
 324              * CALCULATE INTERACTIONS *
 325              **************************/
 326
 327             r10              = _mm_mul_ps(rsq10,rinv10);
 328
 329             /* Compute parameters for interactions between i and j atoms */
 330             qq10             = _mm_mul_ps(iq1,jq0);
 331
 332             /* EWALD ELECTROSTATICS */
 333
 334             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 335             ewrt             = _mm_mul_ps(r10,ewtabscale);
 336             ewitab           = _mm_cvttps_epi32(ewrt);
 337             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 338             ewitab           = _mm_slli_epi32(ewitab,2);
 339             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 340             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 341             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 342             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 343             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 344             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 345             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 346             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 347             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 348
 349             /* Update potential sum for this i atom from the interaction with this j atom. */
 350             velecsum         = _mm_add_ps(velecsum,velec);
 351
 352             fscal            = felec;
 353
 354             /* Calculate temporary vectorial force */
 355             tx               = _mm_mul_ps(fscal,dx10);
 356             ty               = _mm_mul_ps(fscal,dy10);
 357             tz               = _mm_mul_ps(fscal,dz10);
 358
 359             /* Update vectorial force */
 360             fix1             = _mm_add_ps(fix1,tx);
 361             fiy1             = _mm_add_ps(fiy1,ty);
 362             fiz1             = _mm_add_ps(fiz1,tz);
 363
 364             fjx0             = _mm_add_ps(fjx0,tx);
 365             fjy0             = _mm_add_ps(fjy0,ty);
 366             fjz0             = _mm_add_ps(fjz0,tz);
 367
 368             /**************************
 369              * CALCULATE INTERACTIONS *
 370              **************************/
 371
 372             r20              = _mm_mul_ps(rsq20,rinv20);
 373
 374             /* Compute parameters for interactions between i and j atoms */
 375             qq20             = _mm_mul_ps(iq2,jq0);
 376
 377             /* EWALD ELECTROSTATICS */
 378
 379             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 380             ewrt             = _mm_mul_ps(r20,ewtabscale);
 381             ewitab           = _mm_cvttps_epi32(ewrt);
 382             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 383             ewitab           = _mm_slli_epi32(ewitab,2);
 384             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 385             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 386             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 387             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 388             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 389             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 390             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 391             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 392             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 393
 394             /* Update potential sum for this i atom from the interaction with this j atom. */
 395             velecsum         = _mm_add_ps(velecsum,velec);
 396
 397             fscal            = felec;
 398
 399             /* Calculate temporary vectorial force */
 400             tx               = _mm_mul_ps(fscal,dx20);
 401             ty               = _mm_mul_ps(fscal,dy20);
 402             tz               = _mm_mul_ps(fscal,dz20);
 403
 404             /* Update vectorial force */
 405             fix2             = _mm_add_ps(fix2,tx);
 406             fiy2             = _mm_add_ps(fiy2,ty);
 407             fiz2             = _mm_add_ps(fiz2,tz);
 408
 409             fjx0             = _mm_add_ps(fjx0,tx);
 410             fjy0             = _mm_add_ps(fjy0,ty);
 411             fjz0             = _mm_add_ps(fjz0,tz);
 412
 413             fjptrA             = f+j_coord_offsetA;
 414             fjptrB             = f+j_coord_offsetB;
 415             fjptrC             = f+j_coord_offsetC;
 416             fjptrD             = f+j_coord_offsetD;
 417
 418             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 419
 420             /* Inner loop uses 151 flops */
 421         }
 422
 423         if(jidx<j_index_end)
 424         {
 425
 426             /* Get j neighbor index, and coordinate index */
 427             jnrlistA         = jjnr[jidx];
 428             jnrlistB         = jjnr[jidx+1];
 429             jnrlistC         = jjnr[jidx+2];
 430             jnrlistD         = jjnr[jidx+3];
 431             /* Sign of each element will be negative for non-real atoms.
 432              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 433              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 434              */
 435             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 436             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 437             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 438             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 439             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 440             j_coord_offsetA  = DIM*jnrA;
 441             j_coord_offsetB  = DIM*jnrB;
 442             j_coord_offsetC  = DIM*jnrC;
 443             j_coord_offsetD  = DIM*jnrD;
 444
 445             /* load j atom coordinates */
 446             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 447                                               x+j_coord_offsetC,x+j_coord_offsetD,
 448                                               &jx0,&jy0,&jz0);
 449
 450             /* Calculate displacement vector */
 451             dx00             = _mm_sub_ps(ix0,jx0);
 452             dy00             = _mm_sub_ps(iy0,jy0);
 453             dz00             = _mm_sub_ps(iz0,jz0);
 454             dx10             = _mm_sub_ps(ix1,jx0);
 455             dy10             = _mm_sub_ps(iy1,jy0);
 456             dz10             = _mm_sub_ps(iz1,jz0);
 457             dx20             = _mm_sub_ps(ix2,jx0);
 458             dy20             = _mm_sub_ps(iy2,jy0);
 459             dz20             = _mm_sub_ps(iz2,jz0);
 460
 461             /* Calculate squared distance and things based on it */
 462             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 463             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 464             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 465
 466             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 467             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 468             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 469
 470             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 471             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 472             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 473
 474             /* Load parameters for j particles */
 475             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 476                                                               charge+jnrC+0,charge+jnrD+0);
 477             vdwjidx0A        = 2*vdwtype[jnrA+0];
 478             vdwjidx0B        = 2*vdwtype[jnrB+0];
 479             vdwjidx0C        = 2*vdwtype[jnrC+0];
 480             vdwjidx0D        = 2*vdwtype[jnrD+0];
 481
 482             fjx0             = _mm_setzero_ps();
 483             fjy0             = _mm_setzero_ps();
 484             fjz0             = _mm_setzero_ps();
 485
 486             /**************************
 487              * CALCULATE INTERACTIONS *
 488              **************************/
 489
 490             r00              = _mm_mul_ps(rsq00,rinv00);
 491             r00              = _mm_andnot_ps(dummy_mask,r00);
 492
 493             /* Compute parameters for interactions between i and j atoms */
 494             qq00             = _mm_mul_ps(iq0,jq0);
 495             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 496                                          vdwparam+vdwioffset0+vdwjidx0B,
 497                                          vdwparam+vdwioffset0+vdwjidx0C,
 498                                          vdwparam+vdwioffset0+vdwjidx0D,
 499                                          &c6_00,&c12_00);
 500
 501             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 502                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 503                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 504                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 505
 506             /* EWALD ELECTROSTATICS */
 507
 508             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 509             ewrt             = _mm_mul_ps(r00,ewtabscale);
 510             ewitab           = _mm_cvttps_epi32(ewrt);
 511             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 512             ewitab           = _mm_slli_epi32(ewitab,2);
 513             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 514             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 515             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 516             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 517             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 518             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 519             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 520             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 521             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 522
 523             /* Analytical LJ-PME */
 524             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 525             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 526             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 527             exponent         = gmx_simd_exp_r(ewcljrsq);
 528             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 529             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 530             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 531             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 532             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 533             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 534             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 535             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);
 536
 537             /* Update potential sum for this i atom from the interaction with this j atom. */
 538             velec            = _mm_andnot_ps(dummy_mask,velec);
 539             velecsum         = _mm_add_ps(velecsum,velec);
 540             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 541             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 542
 543             fscal            = _mm_add_ps(felec,fvdw);
 544
 545             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 546
 547             /* Calculate temporary vectorial force */
 548             tx               = _mm_mul_ps(fscal,dx00);
 549             ty               = _mm_mul_ps(fscal,dy00);
 550             tz               = _mm_mul_ps(fscal,dz00);
 551
 552             /* Update vectorial force */
 553             fix0             = _mm_add_ps(fix0,tx);
 554             fiy0             = _mm_add_ps(fiy0,ty);
 555             fiz0             = _mm_add_ps(fiz0,tz);
 556
 557             fjx0             = _mm_add_ps(fjx0,tx);
 558             fjy0             = _mm_add_ps(fjy0,ty);
 559             fjz0             = _mm_add_ps(fjz0,tz);
 560
 561             /**************************
 562              * CALCULATE INTERACTIONS *
 563              **************************/
 564
 565             r10              = _mm_mul_ps(rsq10,rinv10);
 566             r10              = _mm_andnot_ps(dummy_mask,r10);
 567
 568             /* Compute parameters for interactions between i and j atoms */
 569             qq10             = _mm_mul_ps(iq1,jq0);
 570
 571             /* EWALD ELECTROSTATICS */
 572
 573             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 574             ewrt             = _mm_mul_ps(r10,ewtabscale);
 575             ewitab           = _mm_cvttps_epi32(ewrt);
 576             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 577             ewitab           = _mm_slli_epi32(ewitab,2);
 578             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 579             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 580             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 581             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 582             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 583             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 584             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 585             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 586             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 587
 588             /* Update potential sum for this i atom from the interaction with this j atom. */
 589             velec            = _mm_andnot_ps(dummy_mask,velec);
 590             velecsum         = _mm_add_ps(velecsum,velec);
 591
 592             fscal            = felec;
 593
 594             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 595
 596             /* Calculate temporary vectorial force */
 597             tx               = _mm_mul_ps(fscal,dx10);
 598             ty               = _mm_mul_ps(fscal,dy10);
 599             tz               = _mm_mul_ps(fscal,dz10);
 600
 601             /* Update vectorial force */
 602             fix1             = _mm_add_ps(fix1,tx);
 603             fiy1             = _mm_add_ps(fiy1,ty);
 604             fiz1             = _mm_add_ps(fiz1,tz);
 605
 606             fjx0             = _mm_add_ps(fjx0,tx);
 607             fjy0             = _mm_add_ps(fjy0,ty);
 608             fjz0             = _mm_add_ps(fjz0,tz);
 609
 610             /**************************
 611              * CALCULATE INTERACTIONS *
 612              **************************/
 613
 614             r20              = _mm_mul_ps(rsq20,rinv20);
 615             r20              = _mm_andnot_ps(dummy_mask,r20);
 616
 617             /* Compute parameters for interactions between i and j atoms */
 618             qq20             = _mm_mul_ps(iq2,jq0);
 619
 620             /* EWALD ELECTROSTATICS */
 621
 622             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 623             ewrt             = _mm_mul_ps(r20,ewtabscale);
 624             ewitab           = _mm_cvttps_epi32(ewrt);
 625             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 626             ewitab           = _mm_slli_epi32(ewitab,2);
 627             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 628             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 629             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 630             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 631             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 632             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 633             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 634             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 635             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 636
 637             /* Update potential sum for this i atom from the interaction with this j atom. */
 638             velec            = _mm_andnot_ps(dummy_mask,velec);
 639             velecsum         = _mm_add_ps(velecsum,velec);
 640
 641             fscal            = felec;
 642
 643             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 644
 645             /* Calculate temporary vectorial force */
 646             tx               = _mm_mul_ps(fscal,dx20);
 647             ty               = _mm_mul_ps(fscal,dy20);
 648             tz               = _mm_mul_ps(fscal,dz20);
 649
 650             /* Update vectorial force */
 651             fix2             = _mm_add_ps(fix2,tx);
 652             fiy2             = _mm_add_ps(fiy2,ty);
 653             fiz2             = _mm_add_ps(fiz2,tz);
 654
 655             fjx0             = _mm_add_ps(fjx0,tx);
 656             fjy0             = _mm_add_ps(fjy0,ty);
 657             fjz0             = _mm_add_ps(fjz0,tz);
 658
 659             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 660             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 661             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 662             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 663
 664             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 665
 666             /* Inner loop uses 154 flops */
 667         }
 668
 669         /* End of innermost loop */
 670
 671         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 672                                               f+i_coord_offset,fshift+i_shift_offset);
 673
 674         ggid                        = gid[iidx];
 675         /* Update potential energies */
 676         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 677         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 678
 679         /* Increment number of inner iterations */
 680         inneriter                  += j_index_end - j_index_start;
 681
 682         /* Outer loop uses 20 flops */
 683     }
 684
 685     /* Increment number of outer iterations */
 686     outeriter        += nri;
 687
 688     /* Update outer/inner flops */
 689
 690     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*154);
 691 }
 692 /*
 693  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_sse4_1_single
 694  * Electrostatics interaction: Ewald
 695  * VdW interaction:            LJEwald
 696  * Geometry:                   Water3-Particle
 697  * Calculate force/pot:        Force
 698  */
 699 void
 700 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_sse4_1_single
 701                     (t_nblist                    * gmx_restrict       nlist,
 702                      rvec                        * gmx_restrict          xx,
 703                      rvec                        * gmx_restrict          ff,
 704                      t_forcerec                  * gmx_restrict          fr,
 705                      t_mdatoms                   * gmx_restrict     mdatoms,
 706                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 707                      t_nrnb                      * gmx_restrict        nrnb)
 708 {
 709     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 710      * just 0 for non-waters.
 711      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 712      * jnr indices corresponding to data put in the four positions in the SIMD register.
 713      */
 714     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 715     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 716     int              jnrA,jnrB,jnrC,jnrD;
 717     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 718     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 719     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 720     real             rcutoff_scalar;
 721     real             *shiftvec,*fshift,*x,*f;
 722     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 723     real             scratch[4*DIM];
 724     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 725     int              vdwioffset0;
 726     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 727     int              vdwioffset1;
 728     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 729     int              vdwioffset2;
 730     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 731     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 732     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 733     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 734     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 735     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 736     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 737     real             *charge;
 738     int              nvdwtype;
 739     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 740     int              *vdwtype;
 741     real             *vdwparam;
 742     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 743     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 744     __m128           c6grid_00;
 745     __m128           c6grid_10;
 746     __m128           c6grid_20;
 747     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 748     real             *vdwgridparam;
 749     __m128           one_half  = _mm_set1_ps(0.5);
 750     __m128           minus_one = _mm_set1_ps(-1.0);
 751     __m128i          ewitab;
 752     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 753     real             *ewtab;
 754     __m128           dummy_mask,cutoff_mask;
 755     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 756     __m128           one     = _mm_set1_ps(1.0);
 757     __m128           two     = _mm_set1_ps(2.0);
 758     x                = xx[0];
 759     f                = ff[0];
 760
 761     nri              = nlist->nri;
 762     iinr             = nlist->iinr;
 763     jindex           = nlist->jindex;
 764     jjnr             = nlist->jjnr;
 765     shiftidx         = nlist->shift;
 766     gid              = nlist->gid;
 767     shiftvec         = fr->shift_vec[0];
 768     fshift           = fr->fshift[0];
 769     facel            = _mm_set1_ps(fr->epsfac);
 770     charge           = mdatoms->chargeA;
 771     nvdwtype         = fr->ntype;
 772     vdwparam         = fr->nbfp;
 773     vdwtype          = mdatoms->typeA;
 774     vdwgridparam     = fr->ljpme_c6grid;
 775     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 776     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 777     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 778
 779     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 780     ewtab            = fr->ic->tabq_coul_F;
 781     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 782     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 783
 784     /* Setup water-specific parameters */
 785     inr              = nlist->iinr[0];
 786     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 787     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 788     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 789     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 790
 791     /* Avoid stupid compiler warnings */
 792     jnrA = jnrB = jnrC = jnrD = 0;
 793     j_coord_offsetA = 0;
 794     j_coord_offsetB = 0;
 795     j_coord_offsetC = 0;
 796     j_coord_offsetD = 0;
 797
 798     outeriter        = 0;
 799     inneriter        = 0;
 800
 801     for(iidx=0;iidx<4*DIM;iidx++)
 802     {
 803         scratch[iidx] = 0.0;
 804     }
 805
 806     /* Start outer loop over neighborlists */
 807     for(iidx=0; iidx<nri; iidx++)
 808     {
 809         /* Load shift vector for this list */
 810         i_shift_offset   = DIM*shiftidx[iidx];
 811
 812         /* Load limits for loop over neighbors */
 813         j_index_start    = jindex[iidx];
 814         j_index_end      = jindex[iidx+1];
 815
 816         /* Get outer coordinate index */
 817         inr              = iinr[iidx];
 818         i_coord_offset   = DIM*inr;
 819
 820         /* Load i particle coords and add shift vector */
 821         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 822                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 823
 824         fix0             = _mm_setzero_ps();
 825         fiy0             = _mm_setzero_ps();
 826         fiz0             = _mm_setzero_ps();
 827         fix1             = _mm_setzero_ps();
 828         fiy1             = _mm_setzero_ps();
 829         fiz1             = _mm_setzero_ps();
 830         fix2             = _mm_setzero_ps();
 831         fiy2             = _mm_setzero_ps();
 832         fiz2             = _mm_setzero_ps();
 833
 834         /* Start inner kernel loop */
 835         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 836         {
 837
 838             /* Get j neighbor index, and coordinate index */
 839             jnrA             = jjnr[jidx];
 840             jnrB             = jjnr[jidx+1];
 841             jnrC             = jjnr[jidx+2];
 842             jnrD             = jjnr[jidx+3];
 843             j_coord_offsetA  = DIM*jnrA;
 844             j_coord_offsetB  = DIM*jnrB;
 845             j_coord_offsetC  = DIM*jnrC;
 846             j_coord_offsetD  = DIM*jnrD;
 847
 848             /* load j atom coordinates */
 849             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 850                                               x+j_coord_offsetC,x+j_coord_offsetD,
 851                                               &jx0,&jy0,&jz0);
 852
 853             /* Calculate displacement vector */
 854             dx00             = _mm_sub_ps(ix0,jx0);
 855             dy00             = _mm_sub_ps(iy0,jy0);
 856             dz00             = _mm_sub_ps(iz0,jz0);
 857             dx10             = _mm_sub_ps(ix1,jx0);
 858             dy10             = _mm_sub_ps(iy1,jy0);
 859             dz10             = _mm_sub_ps(iz1,jz0);
 860             dx20             = _mm_sub_ps(ix2,jx0);
 861             dy20             = _mm_sub_ps(iy2,jy0);
 862             dz20             = _mm_sub_ps(iz2,jz0);
 863
 864             /* Calculate squared distance and things based on it */
 865             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 866             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 867             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 868
 869             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 870             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 871             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 872
 873             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 874             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 875             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 876
 877             /* Load parameters for j particles */
 878             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 879                                                               charge+jnrC+0,charge+jnrD+0);
 880             vdwjidx0A        = 2*vdwtype[jnrA+0];
 881             vdwjidx0B        = 2*vdwtype[jnrB+0];
 882             vdwjidx0C        = 2*vdwtype[jnrC+0];
 883             vdwjidx0D        = 2*vdwtype[jnrD+0];
 884
 885             fjx0             = _mm_setzero_ps();
 886             fjy0             = _mm_setzero_ps();
 887             fjz0             = _mm_setzero_ps();
 888
 889             /**************************
 890              * CALCULATE INTERACTIONS *
 891              **************************/
 892
 893             r00              = _mm_mul_ps(rsq00,rinv00);
 894
 895             /* Compute parameters for interactions between i and j atoms */
 896             qq00             = _mm_mul_ps(iq0,jq0);
 897             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 898                                          vdwparam+vdwioffset0+vdwjidx0B,
 899                                          vdwparam+vdwioffset0+vdwjidx0C,
 900                                          vdwparam+vdwioffset0+vdwjidx0D,
 901                                          &c6_00,&c12_00);
 902
 903             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 904                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 905                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 906                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 907
 908             /* EWALD ELECTROSTATICS */
 909
 910             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 911             ewrt             = _mm_mul_ps(r00,ewtabscale);
 912             ewitab           = _mm_cvttps_epi32(ewrt);
 913             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 914             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 915                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 916                                          &ewtabF,&ewtabFn);
 917             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 918             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 919
 920             /* Analytical LJ-PME */
 921             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 922             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 923             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 924             exponent         = gmx_simd_exp_r(ewcljrsq);
 925             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 926             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 927             /* f6A = 6 * C6grid * (1 - poly) */
 928             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 929             /* f6B = C6grid * exponent * beta^6 */
 930             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 931             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 932             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);
 933
 934             fscal            = _mm_add_ps(felec,fvdw);
 935
 936             /* Calculate temporary vectorial force */
 937             tx               = _mm_mul_ps(fscal,dx00);
 938             ty               = _mm_mul_ps(fscal,dy00);
 939             tz               = _mm_mul_ps(fscal,dz00);
 940
 941             /* Update vectorial force */
 942             fix0             = _mm_add_ps(fix0,tx);
 943             fiy0             = _mm_add_ps(fiy0,ty);
 944             fiz0             = _mm_add_ps(fiz0,tz);
 945
 946             fjx0             = _mm_add_ps(fjx0,tx);
 947             fjy0             = _mm_add_ps(fjy0,ty);
 948             fjz0             = _mm_add_ps(fjz0,tz);
 949
 950             /**************************
 951              * CALCULATE INTERACTIONS *
 952              **************************/
 953
 954             r10              = _mm_mul_ps(rsq10,rinv10);
 955
 956             /* Compute parameters for interactions between i and j atoms */
 957             qq10             = _mm_mul_ps(iq1,jq0);
 958
 959             /* EWALD ELECTROSTATICS */
 960
 961             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 962             ewrt             = _mm_mul_ps(r10,ewtabscale);
 963             ewitab           = _mm_cvttps_epi32(ewrt);
 964             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 965             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 966                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 967                                          &ewtabF,&ewtabFn);
 968             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 969             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 970
 971             fscal            = felec;
 972
 973             /* Calculate temporary vectorial force */
 974             tx               = _mm_mul_ps(fscal,dx10);
 975             ty               = _mm_mul_ps(fscal,dy10);
 976             tz               = _mm_mul_ps(fscal,dz10);
 977
 978             /* Update vectorial force */
 979             fix1             = _mm_add_ps(fix1,tx);
 980             fiy1             = _mm_add_ps(fiy1,ty);
 981             fiz1             = _mm_add_ps(fiz1,tz);
 982
 983             fjx0             = _mm_add_ps(fjx0,tx);
 984             fjy0             = _mm_add_ps(fjy0,ty);
 985             fjz0             = _mm_add_ps(fjz0,tz);
 986
 987             /**************************
 988              * CALCULATE INTERACTIONS *
 989              **************************/
 990
 991             r20              = _mm_mul_ps(rsq20,rinv20);
 992
 993             /* Compute parameters for interactions between i and j atoms */
 994             qq20             = _mm_mul_ps(iq2,jq0);
 995
 996             /* EWALD ELECTROSTATICS */
 997
 998             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 999             ewrt             = _mm_mul_ps(r20,ewtabscale);
1000             ewitab           = _mm_cvttps_epi32(ewrt);
1001             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1002             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1003                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1004                                          &ewtabF,&ewtabFn);
1005             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1006             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1007
1008             fscal            = felec;
1009
1010             /* Calculate temporary vectorial force */
1011             tx               = _mm_mul_ps(fscal,dx20);
1012             ty               = _mm_mul_ps(fscal,dy20);
1013             tz               = _mm_mul_ps(fscal,dz20);
1014
1015             /* Update vectorial force */
1016             fix2             = _mm_add_ps(fix2,tx);
1017             fiy2             = _mm_add_ps(fiy2,ty);
1018             fiz2             = _mm_add_ps(fiz2,tz);
1019
1020             fjx0             = _mm_add_ps(fjx0,tx);
1021             fjy0             = _mm_add_ps(fjy0,ty);
1022             fjz0             = _mm_add_ps(fjz0,tz);
1023
1024             fjptrA             = f+j_coord_offsetA;
1025             fjptrB             = f+j_coord_offsetB;
1026             fjptrC             = f+j_coord_offsetC;
1027             fjptrD             = f+j_coord_offsetD;
1028
1029             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1030
1031             /* Inner loop uses 131 flops */
1032         }
1033
1034         if(jidx<j_index_end)
1035         {
1036
1037             /* Get j neighbor index, and coordinate index */
1038             jnrlistA         = jjnr[jidx];
1039             jnrlistB         = jjnr[jidx+1];
1040             jnrlistC         = jjnr[jidx+2];
1041             jnrlistD         = jjnr[jidx+3];
1042             /* Sign of each element will be negative for non-real atoms.
1043              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1044              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1045              */
1046             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1047             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1048             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1049             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1050             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1051             j_coord_offsetA  = DIM*jnrA;
1052             j_coord_offsetB  = DIM*jnrB;
1053             j_coord_offsetC  = DIM*jnrC;
1054             j_coord_offsetD  = DIM*jnrD;
1055
1056             /* load j atom coordinates */
1057             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1058                                               x+j_coord_offsetC,x+j_coord_offsetD,
1059                                               &jx0,&jy0,&jz0);
1060
1061             /* Calculate displacement vector */
1062             dx00             = _mm_sub_ps(ix0,jx0);
1063             dy00             = _mm_sub_ps(iy0,jy0);
1064             dz00             = _mm_sub_ps(iz0,jz0);
1065             dx10             = _mm_sub_ps(ix1,jx0);
1066             dy10             = _mm_sub_ps(iy1,jy0);
1067             dz10             = _mm_sub_ps(iz1,jz0);
1068             dx20             = _mm_sub_ps(ix2,jx0);
1069             dy20             = _mm_sub_ps(iy2,jy0);
1070             dz20             = _mm_sub_ps(iz2,jz0);
1071
1072             /* Calculate squared distance and things based on it */
1073             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1074             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1075             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1076
1077             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1078             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1079             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1080
1081             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1082             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1083             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1084
1085             /* Load parameters for j particles */
1086             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1087                                                               charge+jnrC+0,charge+jnrD+0);
1088             vdwjidx0A        = 2*vdwtype[jnrA+0];
1089             vdwjidx0B        = 2*vdwtype[jnrB+0];
1090             vdwjidx0C        = 2*vdwtype[jnrC+0];
1091             vdwjidx0D        = 2*vdwtype[jnrD+0];
1092
1093             fjx0             = _mm_setzero_ps();
1094             fjy0             = _mm_setzero_ps();
1095             fjz0             = _mm_setzero_ps();
1096
1097             /**************************
1098              * CALCULATE INTERACTIONS *
1099              **************************/
1100
1101             r00              = _mm_mul_ps(rsq00,rinv00);
1102             r00              = _mm_andnot_ps(dummy_mask,r00);
1103
1104             /* Compute parameters for interactions between i and j atoms */
1105             qq00             = _mm_mul_ps(iq0,jq0);
1106             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1107                                          vdwparam+vdwioffset0+vdwjidx0B,
1108                                          vdwparam+vdwioffset0+vdwjidx0C,
1109                                          vdwparam+vdwioffset0+vdwjidx0D,
1110                                          &c6_00,&c12_00);
1111
1112             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
1113                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
1114                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
1115                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
1116
1117             /* EWALD ELECTROSTATICS */
1118
1119             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1120             ewrt             = _mm_mul_ps(r00,ewtabscale);
1121             ewitab           = _mm_cvttps_epi32(ewrt);
1122             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1123             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1124                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1125                                          &ewtabF,&ewtabFn);
1126             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1127             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1128
1129             /* Analytical LJ-PME */
1130             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1131             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1132             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1133             exponent         = gmx_simd_exp_r(ewcljrsq);
1134             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1135             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1136             /* f6A = 6 * C6grid * (1 - poly) */
1137             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1138             /* f6B = C6grid * exponent * beta^6 */
1139             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1140             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1141             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);
1142
1143             fscal            = _mm_add_ps(felec,fvdw);
1144
1145             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1146
1147             /* Calculate temporary vectorial force */
1148             tx               = _mm_mul_ps(fscal,dx00);
1149             ty               = _mm_mul_ps(fscal,dy00);
1150             tz               = _mm_mul_ps(fscal,dz00);
1151
1152             /* Update vectorial force */
1153             fix0             = _mm_add_ps(fix0,tx);
1154             fiy0             = _mm_add_ps(fiy0,ty);
1155             fiz0             = _mm_add_ps(fiz0,tz);
1156
1157             fjx0             = _mm_add_ps(fjx0,tx);
1158             fjy0             = _mm_add_ps(fjy0,ty);
1159             fjz0             = _mm_add_ps(fjz0,tz);
1160
1161             /**************************
1162              * CALCULATE INTERACTIONS *
1163              **************************/
1164
1165             r10              = _mm_mul_ps(rsq10,rinv10);
1166             r10              = _mm_andnot_ps(dummy_mask,r10);
1167
1168             /* Compute parameters for interactions between i and j atoms */
1169             qq10             = _mm_mul_ps(iq1,jq0);
1170
1171             /* EWALD ELECTROSTATICS */
1172
1173             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1174             ewrt             = _mm_mul_ps(r10,ewtabscale);
1175             ewitab           = _mm_cvttps_epi32(ewrt);
1176             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1177             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1178                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1179                                          &ewtabF,&ewtabFn);
1180             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1181             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1182
1183             fscal            = felec;
1184
1185             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1186
1187             /* Calculate temporary vectorial force */
1188             tx               = _mm_mul_ps(fscal,dx10);
1189             ty               = _mm_mul_ps(fscal,dy10);
1190             tz               = _mm_mul_ps(fscal,dz10);
1191
1192             /* Update vectorial force */
1193             fix1             = _mm_add_ps(fix1,tx);
1194             fiy1             = _mm_add_ps(fiy1,ty);
1195             fiz1             = _mm_add_ps(fiz1,tz);
1196
1197             fjx0             = _mm_add_ps(fjx0,tx);
1198             fjy0             = _mm_add_ps(fjy0,ty);
1199             fjz0             = _mm_add_ps(fjz0,tz);
1200
1201             /**************************
1202              * CALCULATE INTERACTIONS *
1203              **************************/
1204
1205             r20              = _mm_mul_ps(rsq20,rinv20);
1206             r20              = _mm_andnot_ps(dummy_mask,r20);
1207
1208             /* Compute parameters for interactions between i and j atoms */
1209             qq20             = _mm_mul_ps(iq2,jq0);
1210
1211             /* EWALD ELECTROSTATICS */
1212
1213             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1214             ewrt             = _mm_mul_ps(r20,ewtabscale);
1215             ewitab           = _mm_cvttps_epi32(ewrt);
1216             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1217             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1218                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1219                                          &ewtabF,&ewtabFn);
1220             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1221             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1222
1223             fscal            = felec;
1224
1225             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1226
1227             /* Calculate temporary vectorial force */
1228             tx               = _mm_mul_ps(fscal,dx20);
1229             ty               = _mm_mul_ps(fscal,dy20);
1230             tz               = _mm_mul_ps(fscal,dz20);
1231
1232             /* Update vectorial force */
1233             fix2             = _mm_add_ps(fix2,tx);
1234             fiy2             = _mm_add_ps(fiy2,ty);
1235             fiz2             = _mm_add_ps(fiz2,tz);
1236
1237             fjx0             = _mm_add_ps(fjx0,tx);
1238             fjy0             = _mm_add_ps(fjy0,ty);
1239             fjz0             = _mm_add_ps(fjz0,tz);
1240
1241             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1242             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1243             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1244             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1245
1246             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1247
1248             /* Inner loop uses 134 flops */
1249         }
1250
1251         /* End of innermost loop */
1252
1253         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1254                                               f+i_coord_offset,fshift+i_shift_offset);
1255
1256         /* Increment number of inner iterations */
1257         inneriter                  += j_index_end - j_index_start;
1258
1259         /* Outer loop uses 18 flops */
1260     }
1261
1262     /* Increment number of outer iterations */
1263     outeriter        += nri;
1264
1265     /* Update outer/inner flops */
1266
1267     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*134);
1268 }