src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse4_1_single.h"
  50 #include "kernelutil_x86_sse4_1_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_VF_sse4_1_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_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              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  88     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     int              nvdwtype;
  93     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  94     int              *vdwtype;
  95     real             *vdwparam;
  96     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  97     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  98     __m128           c6grid_00;
  99     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 100     real             *vdwgridparam;
 101     __m128           one_half  = _mm_set1_ps(0.5);
 102     __m128           minus_one = _mm_set1_ps(-1.0);
 103     __m128i          ewitab;
 104     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 105     real             *ewtab;
 106     __m128           dummy_mask,cutoff_mask;
 107     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 108     __m128           one     = _mm_set1_ps(1.0);
 109     __m128           two     = _mm_set1_ps(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm_set1_ps(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123     nvdwtype         = fr->ntype;
 124     vdwparam         = fr->nbfp;
 125     vdwtype          = mdatoms->typeA;
 126     vdwgridparam     = fr->ljpme_c6grid;
 127     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 128     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 129     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 130
 131     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 132     ewtab            = fr->ic->tabq_coul_FDV0;
 133     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 134     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 135
 136     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 137     rcutoff_scalar   = fr->rcoulomb;
 138     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 139     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 140
 141     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 142     rvdw             = _mm_set1_ps(fr->rvdw);
 143
 144     /* Avoid stupid compiler warnings */
 145     jnrA = jnrB = jnrC = jnrD = 0;
 146     j_coord_offsetA = 0;
 147     j_coord_offsetB = 0;
 148     j_coord_offsetC = 0;
 149     j_coord_offsetD = 0;
 150
 151     outeriter        = 0;
 152     inneriter        = 0;
 153
 154     for(iidx=0;iidx<4*DIM;iidx++)
 155     {
 156         scratch[iidx] = 0.0;
 157     }
 158
 159     /* Start outer loop over neighborlists */
 160     for(iidx=0; iidx<nri; iidx++)
 161     {
 162         /* Load shift vector for this list */
 163         i_shift_offset   = DIM*shiftidx[iidx];
 164
 165         /* Load limits for loop over neighbors */
 166         j_index_start    = jindex[iidx];
 167         j_index_end      = jindex[iidx+1];
 168
 169         /* Get outer coordinate index */
 170         inr              = iinr[iidx];
 171         i_coord_offset   = DIM*inr;
 172
 173         /* Load i particle coords and add shift vector */
 174         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 175
 176         fix0             = _mm_setzero_ps();
 177         fiy0             = _mm_setzero_ps();
 178         fiz0             = _mm_setzero_ps();
 179
 180         /* Load parameters for i particles */
 181         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 182         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 183
 184         /* Reset potential sums */
 185         velecsum         = _mm_setzero_ps();
 186         vvdwsum          = _mm_setzero_ps();
 187
 188         /* Start inner kernel loop */
 189         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 190         {
 191
 192             /* Get j neighbor index, and coordinate index */
 193             jnrA             = jjnr[jidx];
 194             jnrB             = jjnr[jidx+1];
 195             jnrC             = jjnr[jidx+2];
 196             jnrD             = jjnr[jidx+3];
 197             j_coord_offsetA  = DIM*jnrA;
 198             j_coord_offsetB  = DIM*jnrB;
 199             j_coord_offsetC  = DIM*jnrC;
 200             j_coord_offsetD  = DIM*jnrD;
 201
 202             /* load j atom coordinates */
 203             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 204                                               x+j_coord_offsetC,x+j_coord_offsetD,
 205                                               &jx0,&jy0,&jz0);
 206
 207             /* Calculate displacement vector */
 208             dx00             = _mm_sub_ps(ix0,jx0);
 209             dy00             = _mm_sub_ps(iy0,jy0);
 210             dz00             = _mm_sub_ps(iz0,jz0);
 211
 212             /* Calculate squared distance and things based on it */
 213             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 214
 215             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 216
 217             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 218
 219             /* Load parameters for j particles */
 220             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 221                                                               charge+jnrC+0,charge+jnrD+0);
 222             vdwjidx0A        = 2*vdwtype[jnrA+0];
 223             vdwjidx0B        = 2*vdwtype[jnrB+0];
 224             vdwjidx0C        = 2*vdwtype[jnrC+0];
 225             vdwjidx0D        = 2*vdwtype[jnrD+0];
 226
 227             /**************************
 228              * CALCULATE INTERACTIONS *
 229              **************************/
 230
 231             if (gmx_mm_any_lt(rsq00,rcutoff2))
 232             {
 233
 234             r00              = _mm_mul_ps(rsq00,rinv00);
 235
 236             /* Compute parameters for interactions between i and j atoms */
 237             qq00             = _mm_mul_ps(iq0,jq0);
 238             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 239                                          vdwparam+vdwioffset0+vdwjidx0B,
 240                                          vdwparam+vdwioffset0+vdwjidx0C,
 241                                          vdwparam+vdwioffset0+vdwjidx0D,
 242                                          &c6_00,&c12_00);
 243
 244             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 245                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 246                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 247                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 248
 249             /* EWALD ELECTROSTATICS */
 250
 251             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 252             ewrt             = _mm_mul_ps(r00,ewtabscale);
 253             ewitab           = _mm_cvttps_epi32(ewrt);
 254             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 255             ewitab           = _mm_slli_epi32(ewitab,2);
 256             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 257             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 258             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 259             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 260             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 261             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 262             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 263             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 264             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 265
 266             /* Analytical LJ-PME */
 267             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 268             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 269             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 270             exponent         = gmx_simd_exp_r(ewcljrsq);
 271             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 272             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 273             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 274             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 275             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 276             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))),one_twelfth),
 277                                _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_add_ps(_mm_mul_ps(c6_00,sh_vdw_invrcut6),_mm_mul_ps(c6grid_00,sh_lj_ewald))),one_sixth));
 278             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 279             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);
 280
 281             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 282
 283             /* Update potential sum for this i atom from the interaction with this j atom. */
 284             velec            = _mm_and_ps(velec,cutoff_mask);
 285             velecsum         = _mm_add_ps(velecsum,velec);
 286             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 287             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 288
 289             fscal            = _mm_add_ps(felec,fvdw);
 290
 291             fscal            = _mm_and_ps(fscal,cutoff_mask);
 292
 293             /* Calculate temporary vectorial force */
 294             tx               = _mm_mul_ps(fscal,dx00);
 295             ty               = _mm_mul_ps(fscal,dy00);
 296             tz               = _mm_mul_ps(fscal,dz00);
 297
 298             /* Update vectorial force */
 299             fix0             = _mm_add_ps(fix0,tx);
 300             fiy0             = _mm_add_ps(fiy0,ty);
 301             fiz0             = _mm_add_ps(fiz0,tz);
 302
 303             fjptrA             = f+j_coord_offsetA;
 304             fjptrB             = f+j_coord_offsetB;
 305             fjptrC             = f+j_coord_offsetC;
 306             fjptrD             = f+j_coord_offsetD;
 307             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 308
 309             }
 310
 311             /* Inner loop uses 82 flops */
 312         }
 313
 314         if(jidx<j_index_end)
 315         {
 316
 317             /* Get j neighbor index, and coordinate index */
 318             jnrlistA         = jjnr[jidx];
 319             jnrlistB         = jjnr[jidx+1];
 320             jnrlistC         = jjnr[jidx+2];
 321             jnrlistD         = jjnr[jidx+3];
 322             /* Sign of each element will be negative for non-real atoms.
 323              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 324              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 325              */
 326             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 327             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 328             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 329             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 330             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 331             j_coord_offsetA  = DIM*jnrA;
 332             j_coord_offsetB  = DIM*jnrB;
 333             j_coord_offsetC  = DIM*jnrC;
 334             j_coord_offsetD  = DIM*jnrD;
 335
 336             /* load j atom coordinates */
 337             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 338                                               x+j_coord_offsetC,x+j_coord_offsetD,
 339                                               &jx0,&jy0,&jz0);
 340
 341             /* Calculate displacement vector */
 342             dx00             = _mm_sub_ps(ix0,jx0);
 343             dy00             = _mm_sub_ps(iy0,jy0);
 344             dz00             = _mm_sub_ps(iz0,jz0);
 345
 346             /* Calculate squared distance and things based on it */
 347             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 348
 349             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 350
 351             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 352
 353             /* Load parameters for j particles */
 354             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 355                                                               charge+jnrC+0,charge+jnrD+0);
 356             vdwjidx0A        = 2*vdwtype[jnrA+0];
 357             vdwjidx0B        = 2*vdwtype[jnrB+0];
 358             vdwjidx0C        = 2*vdwtype[jnrC+0];
 359             vdwjidx0D        = 2*vdwtype[jnrD+0];
 360
 361             /**************************
 362              * CALCULATE INTERACTIONS *
 363              **************************/
 364
 365             if (gmx_mm_any_lt(rsq00,rcutoff2))
 366             {
 367
 368             r00              = _mm_mul_ps(rsq00,rinv00);
 369             r00              = _mm_andnot_ps(dummy_mask,r00);
 370
 371             /* Compute parameters for interactions between i and j atoms */
 372             qq00             = _mm_mul_ps(iq0,jq0);
 373             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 374                                          vdwparam+vdwioffset0+vdwjidx0B,
 375                                          vdwparam+vdwioffset0+vdwjidx0C,
 376                                          vdwparam+vdwioffset0+vdwjidx0D,
 377                                          &c6_00,&c12_00);
 378
 379             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 380                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 381                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 382                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 383
 384             /* EWALD ELECTROSTATICS */
 385
 386             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 387             ewrt             = _mm_mul_ps(r00,ewtabscale);
 388             ewitab           = _mm_cvttps_epi32(ewrt);
 389             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 390             ewitab           = _mm_slli_epi32(ewitab,2);
 391             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 392             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 393             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 394             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 395             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 396             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 397             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 398             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 399             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 400
 401             /* Analytical LJ-PME */
 402             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 403             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 404             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 405             exponent         = gmx_simd_exp_r(ewcljrsq);
 406             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 407             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 408             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 409             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 410             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 411             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))),one_twelfth),
 412                                _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_add_ps(_mm_mul_ps(c6_00,sh_vdw_invrcut6),_mm_mul_ps(c6grid_00,sh_lj_ewald))),one_sixth));
 413             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 414             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);
 415
 416             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 417
 418             /* Update potential sum for this i atom from the interaction with this j atom. */
 419             velec            = _mm_and_ps(velec,cutoff_mask);
 420             velec            = _mm_andnot_ps(dummy_mask,velec);
 421             velecsum         = _mm_add_ps(velecsum,velec);
 422             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 423             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 424             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 425
 426             fscal            = _mm_add_ps(felec,fvdw);
 427
 428             fscal            = _mm_and_ps(fscal,cutoff_mask);
 429
 430             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 431
 432             /* Calculate temporary vectorial force */
 433             tx               = _mm_mul_ps(fscal,dx00);
 434             ty               = _mm_mul_ps(fscal,dy00);
 435             tz               = _mm_mul_ps(fscal,dz00);
 436
 437             /* Update vectorial force */
 438             fix0             = _mm_add_ps(fix0,tx);
 439             fiy0             = _mm_add_ps(fiy0,ty);
 440             fiz0             = _mm_add_ps(fiz0,tz);
 441
 442             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 443             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 444             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 445             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 446             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 447
 448             }
 449
 450             /* Inner loop uses 83 flops */
 451         }
 452
 453         /* End of innermost loop */
 454
 455         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 456                                               f+i_coord_offset,fshift+i_shift_offset);
 457
 458         ggid                        = gid[iidx];
 459         /* Update potential energies */
 460         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 461         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 462
 463         /* Increment number of inner iterations */
 464         inneriter                  += j_index_end - j_index_start;
 465
 466         /* Outer loop uses 9 flops */
 467     }
 468
 469     /* Increment number of outer iterations */
 470     outeriter        += nri;
 471
 472     /* Update outer/inner flops */
 473
 474     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*83);
 475 }
 476 /*
 477  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_F_sse4_1_single
 478  * Electrostatics interaction: Ewald
 479  * VdW interaction:            LJEwald
 480  * Geometry:                   Particle-Particle
 481  * Calculate force/pot:        Force
 482  */
 483 void
 484 nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_F_sse4_1_single
 485                     (t_nblist                    * gmx_restrict       nlist,
 486                      rvec                        * gmx_restrict          xx,
 487                      rvec                        * gmx_restrict          ff,
 488                      t_forcerec                  * gmx_restrict          fr,
 489                      t_mdatoms                   * gmx_restrict     mdatoms,
 490                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 491                      t_nrnb                      * gmx_restrict        nrnb)
 492 {
 493     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 494      * just 0 for non-waters.
 495      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 496      * jnr indices corresponding to data put in the four positions in the SIMD register.
 497      */
 498     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 499     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 500     int              jnrA,jnrB,jnrC,jnrD;
 501     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 502     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 503     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 504     real             rcutoff_scalar;
 505     real             *shiftvec,*fshift,*x,*f;
 506     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 507     real             scratch[4*DIM];
 508     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 509     int              vdwioffset0;
 510     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 511     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 512     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 513     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 514     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 515     real             *charge;
 516     int              nvdwtype;
 517     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 518     int              *vdwtype;
 519     real             *vdwparam;
 520     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 521     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 522     __m128           c6grid_00;
 523     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 524     real             *vdwgridparam;
 525     __m128           one_half  = _mm_set1_ps(0.5);
 526     __m128           minus_one = _mm_set1_ps(-1.0);
 527     __m128i          ewitab;
 528     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 529     real             *ewtab;
 530     __m128           dummy_mask,cutoff_mask;
 531     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 532     __m128           one     = _mm_set1_ps(1.0);
 533     __m128           two     = _mm_set1_ps(2.0);
 534     x                = xx[0];
 535     f                = ff[0];
 536
 537     nri              = nlist->nri;
 538     iinr             = nlist->iinr;
 539     jindex           = nlist->jindex;
 540     jjnr             = nlist->jjnr;
 541     shiftidx         = nlist->shift;
 542     gid              = nlist->gid;
 543     shiftvec         = fr->shift_vec[0];
 544     fshift           = fr->fshift[0];
 545     facel            = _mm_set1_ps(fr->epsfac);
 546     charge           = mdatoms->chargeA;
 547     nvdwtype         = fr->ntype;
 548     vdwparam         = fr->nbfp;
 549     vdwtype          = mdatoms->typeA;
 550     vdwgridparam     = fr->ljpme_c6grid;
 551     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 552     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 553     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 554
 555     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 556     ewtab            = fr->ic->tabq_coul_F;
 557     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 558     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 559
 560     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 561     rcutoff_scalar   = fr->rcoulomb;
 562     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 563     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 564
 565     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 566     rvdw             = _mm_set1_ps(fr->rvdw);
 567
 568     /* Avoid stupid compiler warnings */
 569     jnrA = jnrB = jnrC = jnrD = 0;
 570     j_coord_offsetA = 0;
 571     j_coord_offsetB = 0;
 572     j_coord_offsetC = 0;
 573     j_coord_offsetD = 0;
 574
 575     outeriter        = 0;
 576     inneriter        = 0;
 577
 578     for(iidx=0;iidx<4*DIM;iidx++)
 579     {
 580         scratch[iidx] = 0.0;
 581     }
 582
 583     /* Start outer loop over neighborlists */
 584     for(iidx=0; iidx<nri; iidx++)
 585     {
 586         /* Load shift vector for this list */
 587         i_shift_offset   = DIM*shiftidx[iidx];
 588
 589         /* Load limits for loop over neighbors */
 590         j_index_start    = jindex[iidx];
 591         j_index_end      = jindex[iidx+1];
 592
 593         /* Get outer coordinate index */
 594         inr              = iinr[iidx];
 595         i_coord_offset   = DIM*inr;
 596
 597         /* Load i particle coords and add shift vector */
 598         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 599
 600         fix0             = _mm_setzero_ps();
 601         fiy0             = _mm_setzero_ps();
 602         fiz0             = _mm_setzero_ps();
 603
 604         /* Load parameters for i particles */
 605         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 606         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 607
 608         /* Start inner kernel loop */
 609         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 610         {
 611
 612             /* Get j neighbor index, and coordinate index */
 613             jnrA             = jjnr[jidx];
 614             jnrB             = jjnr[jidx+1];
 615             jnrC             = jjnr[jidx+2];
 616             jnrD             = jjnr[jidx+3];
 617             j_coord_offsetA  = DIM*jnrA;
 618             j_coord_offsetB  = DIM*jnrB;
 619             j_coord_offsetC  = DIM*jnrC;
 620             j_coord_offsetD  = DIM*jnrD;
 621
 622             /* load j atom coordinates */
 623             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 624                                               x+j_coord_offsetC,x+j_coord_offsetD,
 625                                               &jx0,&jy0,&jz0);
 626
 627             /* Calculate displacement vector */
 628             dx00             = _mm_sub_ps(ix0,jx0);
 629             dy00             = _mm_sub_ps(iy0,jy0);
 630             dz00             = _mm_sub_ps(iz0,jz0);
 631
 632             /* Calculate squared distance and things based on it */
 633             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 634
 635             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 636
 637             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 638
 639             /* Load parameters for j particles */
 640             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 641                                                               charge+jnrC+0,charge+jnrD+0);
 642             vdwjidx0A        = 2*vdwtype[jnrA+0];
 643             vdwjidx0B        = 2*vdwtype[jnrB+0];
 644             vdwjidx0C        = 2*vdwtype[jnrC+0];
 645             vdwjidx0D        = 2*vdwtype[jnrD+0];
 646
 647             /**************************
 648              * CALCULATE INTERACTIONS *
 649              **************************/
 650
 651             if (gmx_mm_any_lt(rsq00,rcutoff2))
 652             {
 653
 654             r00              = _mm_mul_ps(rsq00,rinv00);
 655
 656             /* Compute parameters for interactions between i and j atoms */
 657             qq00             = _mm_mul_ps(iq0,jq0);
 658             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 659                                          vdwparam+vdwioffset0+vdwjidx0B,
 660                                          vdwparam+vdwioffset0+vdwjidx0C,
 661                                          vdwparam+vdwioffset0+vdwjidx0D,
 662                                          &c6_00,&c12_00);
 663
 664             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 665                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 666                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 667                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 668
 669             /* EWALD ELECTROSTATICS */
 670
 671             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 672             ewrt             = _mm_mul_ps(r00,ewtabscale);
 673             ewitab           = _mm_cvttps_epi32(ewrt);
 674             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 675             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 676                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 677                                          &ewtabF,&ewtabFn);
 678             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 679             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 680
 681             /* Analytical LJ-PME */
 682             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 683             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 684             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 685             exponent         = gmx_simd_exp_r(ewcljrsq);
 686             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 687             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 688             /* f6A = 6 * C6grid * (1 - poly) */
 689             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 690             /* f6B = C6grid * exponent * beta^6 */
 691             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 692             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 693             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);
 694
 695             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 696
 697             fscal            = _mm_add_ps(felec,fvdw);
 698
 699             fscal            = _mm_and_ps(fscal,cutoff_mask);
 700
 701             /* Calculate temporary vectorial force */
 702             tx               = _mm_mul_ps(fscal,dx00);
 703             ty               = _mm_mul_ps(fscal,dy00);
 704             tz               = _mm_mul_ps(fscal,dz00);
 705
 706             /* Update vectorial force */
 707             fix0             = _mm_add_ps(fix0,tx);
 708             fiy0             = _mm_add_ps(fiy0,ty);
 709             fiz0             = _mm_add_ps(fiz0,tz);
 710
 711             fjptrA             = f+j_coord_offsetA;
 712             fjptrB             = f+j_coord_offsetB;
 713             fjptrC             = f+j_coord_offsetC;
 714             fjptrD             = f+j_coord_offsetD;
 715             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 716
 717             }
 718
 719             /* Inner loop uses 62 flops */
 720         }
 721
 722         if(jidx<j_index_end)
 723         {
 724
 725             /* Get j neighbor index, and coordinate index */
 726             jnrlistA         = jjnr[jidx];
 727             jnrlistB         = jjnr[jidx+1];
 728             jnrlistC         = jjnr[jidx+2];
 729             jnrlistD         = jjnr[jidx+3];
 730             /* Sign of each element will be negative for non-real atoms.
 731              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 732              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 733              */
 734             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 735             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 736             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 737             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 738             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 739             j_coord_offsetA  = DIM*jnrA;
 740             j_coord_offsetB  = DIM*jnrB;
 741             j_coord_offsetC  = DIM*jnrC;
 742             j_coord_offsetD  = DIM*jnrD;
 743
 744             /* load j atom coordinates */
 745             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 746                                               x+j_coord_offsetC,x+j_coord_offsetD,
 747                                               &jx0,&jy0,&jz0);
 748
 749             /* Calculate displacement vector */
 750             dx00             = _mm_sub_ps(ix0,jx0);
 751             dy00             = _mm_sub_ps(iy0,jy0);
 752             dz00             = _mm_sub_ps(iz0,jz0);
 753
 754             /* Calculate squared distance and things based on it */
 755             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 756
 757             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 758
 759             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 760
 761             /* Load parameters for j particles */
 762             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 763                                                               charge+jnrC+0,charge+jnrD+0);
 764             vdwjidx0A        = 2*vdwtype[jnrA+0];
 765             vdwjidx0B        = 2*vdwtype[jnrB+0];
 766             vdwjidx0C        = 2*vdwtype[jnrC+0];
 767             vdwjidx0D        = 2*vdwtype[jnrD+0];
 768
 769             /**************************
 770              * CALCULATE INTERACTIONS *
 771              **************************/
 772
 773             if (gmx_mm_any_lt(rsq00,rcutoff2))
 774             {
 775
 776             r00              = _mm_mul_ps(rsq00,rinv00);
 777             r00              = _mm_andnot_ps(dummy_mask,r00);
 778
 779             /* Compute parameters for interactions between i and j atoms */
 780             qq00             = _mm_mul_ps(iq0,jq0);
 781             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 782                                          vdwparam+vdwioffset0+vdwjidx0B,
 783                                          vdwparam+vdwioffset0+vdwjidx0C,
 784                                          vdwparam+vdwioffset0+vdwjidx0D,
 785                                          &c6_00,&c12_00);
 786
 787             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 788                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 789                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 790                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 791
 792             /* EWALD ELECTROSTATICS */
 793
 794             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 795             ewrt             = _mm_mul_ps(r00,ewtabscale);
 796             ewitab           = _mm_cvttps_epi32(ewrt);
 797             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 798             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 799                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 800                                          &ewtabF,&ewtabFn);
 801             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 802             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 803
 804             /* Analytical LJ-PME */
 805             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 806             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 807             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 808             exponent         = gmx_simd_exp_r(ewcljrsq);
 809             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 810             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 811             /* f6A = 6 * C6grid * (1 - poly) */
 812             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 813             /* f6B = C6grid * exponent * beta^6 */
 814             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 815             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 816             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);
 817
 818             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 819
 820             fscal            = _mm_add_ps(felec,fvdw);
 821
 822             fscal            = _mm_and_ps(fscal,cutoff_mask);
 823
 824             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 825
 826             /* Calculate temporary vectorial force */
 827             tx               = _mm_mul_ps(fscal,dx00);
 828             ty               = _mm_mul_ps(fscal,dy00);
 829             tz               = _mm_mul_ps(fscal,dz00);
 830
 831             /* Update vectorial force */
 832             fix0             = _mm_add_ps(fix0,tx);
 833             fiy0             = _mm_add_ps(fiy0,ty);
 834             fiz0             = _mm_add_ps(fiz0,tz);
 835
 836             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 837             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 838             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 839             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 840             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 841
 842             }
 843
 844             /* Inner loop uses 63 flops */
 845         }
 846
 847         /* End of innermost loop */
 848
 849         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 850                                               f+i_coord_offset,fshift+i_shift_offset);
 851
 852         /* Increment number of inner iterations */
 853         inneriter                  += j_index_end - j_index_start;
 854
 855         /* Outer loop uses 7 flops */
 856     }
 857
 858     /* Increment number of outer iterations */
 859     outeriter        += nri;
 860
 861     /* Update outer/inner flops */
 862
 863     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*63);
 864 }