src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_sse2_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_single.h"
  48 #include "kernelutil_x86_sse2_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_VF_sse2_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LJEwald
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_VF_sse2_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  86     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  89     real             *charge;
  90     int              nvdwtype;
  91     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  92     int              *vdwtype;
  93     real             *vdwparam;
  94     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  95     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  96     __m128           c6grid_00;
  97     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
  98     real             *vdwgridparam;
  99     __m128           one_half = _mm_set1_ps(0.5);
 100     __m128           minus_one = _mm_set1_ps(-1.0);
 101     __m128i          ewitab;
 102     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 103     real             *ewtab;
 104     __m128           dummy_mask,cutoff_mask;
 105     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 106     __m128           one     = _mm_set1_ps(1.0);
 107     __m128           two     = _mm_set1_ps(2.0);
 108     x                = xx[0];
 109     f                = ff[0];
 110
 111     nri              = nlist->nri;
 112     iinr             = nlist->iinr;
 113     jindex           = nlist->jindex;
 114     jjnr             = nlist->jjnr;
 115     shiftidx         = nlist->shift;
 116     gid              = nlist->gid;
 117     shiftvec         = fr->shift_vec[0];
 118     fshift           = fr->fshift[0];
 119     facel            = _mm_set1_ps(fr->epsfac);
 120     charge           = mdatoms->chargeA;
 121     nvdwtype         = fr->ntype;
 122     vdwparam         = fr->nbfp;
 123     vdwtype          = mdatoms->typeA;
 124     vdwgridparam     = fr->ljpme_c6grid;
 125     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 126     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 127     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 128
 129     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 130     ewtab            = fr->ic->tabq_coul_FDV0;
 131     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 132     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 133
 134     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 135     rcutoff_scalar   = fr->rcoulomb;
 136     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 137     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 138
 139     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 140     rvdw             = _mm_set1_ps(fr->rvdw);
 141
 142     /* Avoid stupid compiler warnings */
 143     jnrA = jnrB = jnrC = jnrD = 0;
 144     j_coord_offsetA = 0;
 145     j_coord_offsetB = 0;
 146     j_coord_offsetC = 0;
 147     j_coord_offsetD = 0;
 148
 149     outeriter        = 0;
 150     inneriter        = 0;
 151
 152     for(iidx=0;iidx<4*DIM;iidx++)
 153     {
 154         scratch[iidx] = 0.0;
 155     }
 156
 157     /* Start outer loop over neighborlists */
 158     for(iidx=0; iidx<nri; iidx++)
 159     {
 160         /* Load shift vector for this list */
 161         i_shift_offset   = DIM*shiftidx[iidx];
 162
 163         /* Load limits for loop over neighbors */
 164         j_index_start    = jindex[iidx];
 165         j_index_end      = jindex[iidx+1];
 166
 167         /* Get outer coordinate index */
 168         inr              = iinr[iidx];
 169         i_coord_offset   = DIM*inr;
 170
 171         /* Load i particle coords and add shift vector */
 172         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 173
 174         fix0             = _mm_setzero_ps();
 175         fiy0             = _mm_setzero_ps();
 176         fiz0             = _mm_setzero_ps();
 177
 178         /* Load parameters for i particles */
 179         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 180         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 181
 182         /* Reset potential sums */
 183         velecsum         = _mm_setzero_ps();
 184         vvdwsum          = _mm_setzero_ps();
 185
 186         /* Start inner kernel loop */
 187         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 188         {
 189
 190             /* Get j neighbor index, and coordinate index */
 191             jnrA             = jjnr[jidx];
 192             jnrB             = jjnr[jidx+1];
 193             jnrC             = jjnr[jidx+2];
 194             jnrD             = jjnr[jidx+3];
 195             j_coord_offsetA  = DIM*jnrA;
 196             j_coord_offsetB  = DIM*jnrB;
 197             j_coord_offsetC  = DIM*jnrC;
 198             j_coord_offsetD  = DIM*jnrD;
 199
 200             /* load j atom coordinates */
 201             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 202                                               x+j_coord_offsetC,x+j_coord_offsetD,
 203                                               &jx0,&jy0,&jz0);
 204
 205             /* Calculate displacement vector */
 206             dx00             = _mm_sub_ps(ix0,jx0);
 207             dy00             = _mm_sub_ps(iy0,jy0);
 208             dz00             = _mm_sub_ps(iz0,jz0);
 209
 210             /* Calculate squared distance and things based on it */
 211             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 212
 213             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 214
 215             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 216
 217             /* Load parameters for j particles */
 218             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 219                                                               charge+jnrC+0,charge+jnrD+0);
 220             vdwjidx0A        = 2*vdwtype[jnrA+0];
 221             vdwjidx0B        = 2*vdwtype[jnrB+0];
 222             vdwjidx0C        = 2*vdwtype[jnrC+0];
 223             vdwjidx0D        = 2*vdwtype[jnrD+0];
 224
 225             /**************************
 226              * CALCULATE INTERACTIONS *
 227              **************************/
 228
 229             if (gmx_mm_any_lt(rsq00,rcutoff2))
 230             {
 231
 232             r00              = _mm_mul_ps(rsq00,rinv00);
 233
 234             /* Compute parameters for interactions between i and j atoms */
 235             qq00             = _mm_mul_ps(iq0,jq0);
 236             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 237                                          vdwparam+vdwioffset0+vdwjidx0B,
 238                                          vdwparam+vdwioffset0+vdwjidx0C,
 239                                          vdwparam+vdwioffset0+vdwjidx0D,
 240                                          &c6_00,&c12_00);
 241             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 242                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 243                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 244                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 245
 246             /* EWALD ELECTROSTATICS */
 247
 248             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 249             ewrt             = _mm_mul_ps(r00,ewtabscale);
 250             ewitab           = _mm_cvttps_epi32(ewrt);
 251             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 252             ewitab           = _mm_slli_epi32(ewitab,2);
 253             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 254             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 255             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 256             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 257             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 258             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 259             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 260             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 261             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 262
 263             /* Analytical LJ-PME */
 264             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 265             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 266             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 267             exponent         = gmx_simd_exp_r(ewcljrsq);
 268             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 269             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 270             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 271             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 272             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 273             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) ,
 274                                           _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));
 275             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 276             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);
 277
 278             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 279
 280             /* Update potential sum for this i atom from the interaction with this j atom. */
 281             velec            = _mm_and_ps(velec,cutoff_mask);
 282             velecsum         = _mm_add_ps(velecsum,velec);
 283             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 284             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 285
 286             fscal            = _mm_add_ps(felec,fvdw);
 287
 288             fscal            = _mm_and_ps(fscal,cutoff_mask);
 289
 290             /* Calculate temporary vectorial force */
 291             tx               = _mm_mul_ps(fscal,dx00);
 292             ty               = _mm_mul_ps(fscal,dy00);
 293             tz               = _mm_mul_ps(fscal,dz00);
 294
 295             /* Update vectorial force */
 296             fix0             = _mm_add_ps(fix0,tx);
 297             fiy0             = _mm_add_ps(fiy0,ty);
 298             fiz0             = _mm_add_ps(fiz0,tz);
 299
 300             fjptrA             = f+j_coord_offsetA;
 301             fjptrB             = f+j_coord_offsetB;
 302             fjptrC             = f+j_coord_offsetC;
 303             fjptrD             = f+j_coord_offsetD;
 304             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 305
 306             }
 307
 308             /* Inner loop uses 82 flops */
 309         }
 310
 311         if(jidx<j_index_end)
 312         {
 313
 314             /* Get j neighbor index, and coordinate index */
 315             jnrlistA         = jjnr[jidx];
 316             jnrlistB         = jjnr[jidx+1];
 317             jnrlistC         = jjnr[jidx+2];
 318             jnrlistD         = jjnr[jidx+3];
 319             /* Sign of each element will be negative for non-real atoms.
 320              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 321              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 322              */
 323             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 324             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 325             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 326             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 327             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 328             j_coord_offsetA  = DIM*jnrA;
 329             j_coord_offsetB  = DIM*jnrB;
 330             j_coord_offsetC  = DIM*jnrC;
 331             j_coord_offsetD  = DIM*jnrD;
 332
 333             /* load j atom coordinates */
 334             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 335                                               x+j_coord_offsetC,x+j_coord_offsetD,
 336                                               &jx0,&jy0,&jz0);
 337
 338             /* Calculate displacement vector */
 339             dx00             = _mm_sub_ps(ix0,jx0);
 340             dy00             = _mm_sub_ps(iy0,jy0);
 341             dz00             = _mm_sub_ps(iz0,jz0);
 342
 343             /* Calculate squared distance and things based on it */
 344             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 345
 346             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 347
 348             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 349
 350             /* Load parameters for j particles */
 351             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 352                                                               charge+jnrC+0,charge+jnrD+0);
 353             vdwjidx0A        = 2*vdwtype[jnrA+0];
 354             vdwjidx0B        = 2*vdwtype[jnrB+0];
 355             vdwjidx0C        = 2*vdwtype[jnrC+0];
 356             vdwjidx0D        = 2*vdwtype[jnrD+0];
 357
 358             /**************************
 359              * CALCULATE INTERACTIONS *
 360              **************************/
 361
 362             if (gmx_mm_any_lt(rsq00,rcutoff2))
 363             {
 364
 365             r00              = _mm_mul_ps(rsq00,rinv00);
 366             r00              = _mm_andnot_ps(dummy_mask,r00);
 367
 368             /* Compute parameters for interactions between i and j atoms */
 369             qq00             = _mm_mul_ps(iq0,jq0);
 370             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 371                                          vdwparam+vdwioffset0+vdwjidx0B,
 372                                          vdwparam+vdwioffset0+vdwjidx0C,
 373                                          vdwparam+vdwioffset0+vdwjidx0D,
 374                                          &c6_00,&c12_00);
 375             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 376                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 377                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 378                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 379
 380             /* EWALD ELECTROSTATICS */
 381
 382             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 383             ewrt             = _mm_mul_ps(r00,ewtabscale);
 384             ewitab           = _mm_cvttps_epi32(ewrt);
 385             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 386             ewitab           = _mm_slli_epi32(ewitab,2);
 387             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 388             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 389             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 390             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 391             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 392             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 393             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 394             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 395             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 396
 397             /* Analytical LJ-PME */
 398             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 399             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 400             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 401             exponent         = gmx_simd_exp_r(ewcljrsq);
 402             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 403             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 404             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 405             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 406             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 407             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) ,
 408                                           _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));
 409             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 410             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);
 411
 412             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 413
 414             /* Update potential sum for this i atom from the interaction with this j atom. */
 415             velec            = _mm_and_ps(velec,cutoff_mask);
 416             velec            = _mm_andnot_ps(dummy_mask,velec);
 417             velecsum         = _mm_add_ps(velecsum,velec);
 418             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 419             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 420             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 421
 422             fscal            = _mm_add_ps(felec,fvdw);
 423
 424             fscal            = _mm_and_ps(fscal,cutoff_mask);
 425
 426             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 427
 428             /* Calculate temporary vectorial force */
 429             tx               = _mm_mul_ps(fscal,dx00);
 430             ty               = _mm_mul_ps(fscal,dy00);
 431             tz               = _mm_mul_ps(fscal,dz00);
 432
 433             /* Update vectorial force */
 434             fix0             = _mm_add_ps(fix0,tx);
 435             fiy0             = _mm_add_ps(fiy0,ty);
 436             fiz0             = _mm_add_ps(fiz0,tz);
 437
 438             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 439             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 440             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 441             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 442             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 443
 444             }
 445
 446             /* Inner loop uses 83 flops */
 447         }
 448
 449         /* End of innermost loop */
 450
 451         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 452                                               f+i_coord_offset,fshift+i_shift_offset);
 453
 454         ggid                        = gid[iidx];
 455         /* Update potential energies */
 456         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 457         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 458
 459         /* Increment number of inner iterations */
 460         inneriter                  += j_index_end - j_index_start;
 461
 462         /* Outer loop uses 9 flops */
 463     }
 464
 465     /* Increment number of outer iterations */
 466     outeriter        += nri;
 467
 468     /* Update outer/inner flops */
 469
 470     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*83);
 471 }
 472 /*
 473  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_F_sse2_single
 474  * Electrostatics interaction: Ewald
 475  * VdW interaction:            LJEwald
 476  * Geometry:                   Particle-Particle
 477  * Calculate force/pot:        Force
 478  */
 479 void
 480 nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_F_sse2_single
 481                     (t_nblist                    * gmx_restrict       nlist,
 482                      rvec                        * gmx_restrict          xx,
 483                      rvec                        * gmx_restrict          ff,
 484                      t_forcerec                  * gmx_restrict          fr,
 485                      t_mdatoms                   * gmx_restrict     mdatoms,
 486                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 487                      t_nrnb                      * gmx_restrict        nrnb)
 488 {
 489     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 490      * just 0 for non-waters.
 491      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 492      * jnr indices corresponding to data put in the four positions in the SIMD register.
 493      */
 494     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 495     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 496     int              jnrA,jnrB,jnrC,jnrD;
 497     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 498     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 499     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 500     real             rcutoff_scalar;
 501     real             *shiftvec,*fshift,*x,*f;
 502     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 503     real             scratch[4*DIM];
 504     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 505     int              vdwioffset0;
 506     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 507     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 508     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 509     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 510     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 511     real             *charge;
 512     int              nvdwtype;
 513     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 514     int              *vdwtype;
 515     real             *vdwparam;
 516     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 517     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 518     __m128           c6grid_00;
 519     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 520     real             *vdwgridparam;
 521     __m128           one_half = _mm_set1_ps(0.5);
 522     __m128           minus_one = _mm_set1_ps(-1.0);
 523     __m128i          ewitab;
 524     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 525     real             *ewtab;
 526     __m128           dummy_mask,cutoff_mask;
 527     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 528     __m128           one     = _mm_set1_ps(1.0);
 529     __m128           two     = _mm_set1_ps(2.0);
 530     x                = xx[0];
 531     f                = ff[0];
 532
 533     nri              = nlist->nri;
 534     iinr             = nlist->iinr;
 535     jindex           = nlist->jindex;
 536     jjnr             = nlist->jjnr;
 537     shiftidx         = nlist->shift;
 538     gid              = nlist->gid;
 539     shiftvec         = fr->shift_vec[0];
 540     fshift           = fr->fshift[0];
 541     facel            = _mm_set1_ps(fr->epsfac);
 542     charge           = mdatoms->chargeA;
 543     nvdwtype         = fr->ntype;
 544     vdwparam         = fr->nbfp;
 545     vdwtype          = mdatoms->typeA;
 546     vdwgridparam     = fr->ljpme_c6grid;
 547     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 548     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 549     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 550
 551     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 552     ewtab            = fr->ic->tabq_coul_F;
 553     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 554     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 555
 556     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 557     rcutoff_scalar   = fr->rcoulomb;
 558     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 559     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 560
 561     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 562     rvdw             = _mm_set1_ps(fr->rvdw);
 563
 564     /* Avoid stupid compiler warnings */
 565     jnrA = jnrB = jnrC = jnrD = 0;
 566     j_coord_offsetA = 0;
 567     j_coord_offsetB = 0;
 568     j_coord_offsetC = 0;
 569     j_coord_offsetD = 0;
 570
 571     outeriter        = 0;
 572     inneriter        = 0;
 573
 574     for(iidx=0;iidx<4*DIM;iidx++)
 575     {
 576         scratch[iidx] = 0.0;
 577     }
 578
 579     /* Start outer loop over neighborlists */
 580     for(iidx=0; iidx<nri; iidx++)
 581     {
 582         /* Load shift vector for this list */
 583         i_shift_offset   = DIM*shiftidx[iidx];
 584
 585         /* Load limits for loop over neighbors */
 586         j_index_start    = jindex[iidx];
 587         j_index_end      = jindex[iidx+1];
 588
 589         /* Get outer coordinate index */
 590         inr              = iinr[iidx];
 591         i_coord_offset   = DIM*inr;
 592
 593         /* Load i particle coords and add shift vector */
 594         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 595
 596         fix0             = _mm_setzero_ps();
 597         fiy0             = _mm_setzero_ps();
 598         fiz0             = _mm_setzero_ps();
 599
 600         /* Load parameters for i particles */
 601         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 602         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 603
 604         /* Start inner kernel loop */
 605         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 606         {
 607
 608             /* Get j neighbor index, and coordinate index */
 609             jnrA             = jjnr[jidx];
 610             jnrB             = jjnr[jidx+1];
 611             jnrC             = jjnr[jidx+2];
 612             jnrD             = jjnr[jidx+3];
 613             j_coord_offsetA  = DIM*jnrA;
 614             j_coord_offsetB  = DIM*jnrB;
 615             j_coord_offsetC  = DIM*jnrC;
 616             j_coord_offsetD  = DIM*jnrD;
 617
 618             /* load j atom coordinates */
 619             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 620                                               x+j_coord_offsetC,x+j_coord_offsetD,
 621                                               &jx0,&jy0,&jz0);
 622
 623             /* Calculate displacement vector */
 624             dx00             = _mm_sub_ps(ix0,jx0);
 625             dy00             = _mm_sub_ps(iy0,jy0);
 626             dz00             = _mm_sub_ps(iz0,jz0);
 627
 628             /* Calculate squared distance and things based on it */
 629             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 630
 631             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 632
 633             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 634
 635             /* Load parameters for j particles */
 636             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 637                                                               charge+jnrC+0,charge+jnrD+0);
 638             vdwjidx0A        = 2*vdwtype[jnrA+0];
 639             vdwjidx0B        = 2*vdwtype[jnrB+0];
 640             vdwjidx0C        = 2*vdwtype[jnrC+0];
 641             vdwjidx0D        = 2*vdwtype[jnrD+0];
 642
 643             /**************************
 644              * CALCULATE INTERACTIONS *
 645              **************************/
 646
 647             if (gmx_mm_any_lt(rsq00,rcutoff2))
 648             {
 649
 650             r00              = _mm_mul_ps(rsq00,rinv00);
 651
 652             /* Compute parameters for interactions between i and j atoms */
 653             qq00             = _mm_mul_ps(iq0,jq0);
 654             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 655                                          vdwparam+vdwioffset0+vdwjidx0B,
 656                                          vdwparam+vdwioffset0+vdwjidx0C,
 657                                          vdwparam+vdwioffset0+vdwjidx0D,
 658                                          &c6_00,&c12_00);
 659             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 660                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 661                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 662                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 663
 664             /* EWALD ELECTROSTATICS */
 665
 666             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 667             ewrt             = _mm_mul_ps(r00,ewtabscale);
 668             ewitab           = _mm_cvttps_epi32(ewrt);
 669             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 670             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 671                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 672                                          &ewtabF,&ewtabFn);
 673             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 674             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 675
 676             /* Analytical LJ-PME */
 677             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 678             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 679             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 680             exponent         = gmx_simd_exp_r(ewcljrsq);
 681             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 682             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 683             /* f6A = 6 * C6grid * (1 - poly) */
 684             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 685             /* f6B = C6grid * exponent * beta^6 */
 686             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 687             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 688             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);
 689
 690             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 691
 692             fscal            = _mm_add_ps(felec,fvdw);
 693
 694             fscal            = _mm_and_ps(fscal,cutoff_mask);
 695
 696             /* Calculate temporary vectorial force */
 697             tx               = _mm_mul_ps(fscal,dx00);
 698             ty               = _mm_mul_ps(fscal,dy00);
 699             tz               = _mm_mul_ps(fscal,dz00);
 700
 701             /* Update vectorial force */
 702             fix0             = _mm_add_ps(fix0,tx);
 703             fiy0             = _mm_add_ps(fiy0,ty);
 704             fiz0             = _mm_add_ps(fiz0,tz);
 705
 706             fjptrA             = f+j_coord_offsetA;
 707             fjptrB             = f+j_coord_offsetB;
 708             fjptrC             = f+j_coord_offsetC;
 709             fjptrD             = f+j_coord_offsetD;
 710             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 711
 712             }
 713
 714             /* Inner loop uses 62 flops */
 715         }
 716
 717         if(jidx<j_index_end)
 718         {
 719
 720             /* Get j neighbor index, and coordinate index */
 721             jnrlistA         = jjnr[jidx];
 722             jnrlistB         = jjnr[jidx+1];
 723             jnrlistC         = jjnr[jidx+2];
 724             jnrlistD         = jjnr[jidx+3];
 725             /* Sign of each element will be negative for non-real atoms.
 726              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 727              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 728              */
 729             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 730             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 731             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 732             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 733             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 734             j_coord_offsetA  = DIM*jnrA;
 735             j_coord_offsetB  = DIM*jnrB;
 736             j_coord_offsetC  = DIM*jnrC;
 737             j_coord_offsetD  = DIM*jnrD;
 738
 739             /* load j atom coordinates */
 740             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 741                                               x+j_coord_offsetC,x+j_coord_offsetD,
 742                                               &jx0,&jy0,&jz0);
 743
 744             /* Calculate displacement vector */
 745             dx00             = _mm_sub_ps(ix0,jx0);
 746             dy00             = _mm_sub_ps(iy0,jy0);
 747             dz00             = _mm_sub_ps(iz0,jz0);
 748
 749             /* Calculate squared distance and things based on it */
 750             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 751
 752             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 753
 754             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 755
 756             /* Load parameters for j particles */
 757             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 758                                                               charge+jnrC+0,charge+jnrD+0);
 759             vdwjidx0A        = 2*vdwtype[jnrA+0];
 760             vdwjidx0B        = 2*vdwtype[jnrB+0];
 761             vdwjidx0C        = 2*vdwtype[jnrC+0];
 762             vdwjidx0D        = 2*vdwtype[jnrD+0];
 763
 764             /**************************
 765              * CALCULATE INTERACTIONS *
 766              **************************/
 767
 768             if (gmx_mm_any_lt(rsq00,rcutoff2))
 769             {
 770
 771             r00              = _mm_mul_ps(rsq00,rinv00);
 772             r00              = _mm_andnot_ps(dummy_mask,r00);
 773
 774             /* Compute parameters for interactions between i and j atoms */
 775             qq00             = _mm_mul_ps(iq0,jq0);
 776             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 777                                          vdwparam+vdwioffset0+vdwjidx0B,
 778                                          vdwparam+vdwioffset0+vdwjidx0C,
 779                                          vdwparam+vdwioffset0+vdwjidx0D,
 780                                          &c6_00,&c12_00);
 781             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 782                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 783                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 784                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 785
 786             /* EWALD ELECTROSTATICS */
 787
 788             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 789             ewrt             = _mm_mul_ps(r00,ewtabscale);
 790             ewitab           = _mm_cvttps_epi32(ewrt);
 791             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 792             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 793                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 794                                          &ewtabF,&ewtabFn);
 795             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 796             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 797
 798             /* Analytical LJ-PME */
 799             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 800             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 801             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 802             exponent         = gmx_simd_exp_r(ewcljrsq);
 803             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 804             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 805             /* f6A = 6 * C6grid * (1 - poly) */
 806             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 807             /* f6B = C6grid * exponent * beta^6 */
 808             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 809             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 810             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);
 811
 812             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 813
 814             fscal            = _mm_add_ps(felec,fvdw);
 815
 816             fscal            = _mm_and_ps(fscal,cutoff_mask);
 817
 818             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 819
 820             /* Calculate temporary vectorial force */
 821             tx               = _mm_mul_ps(fscal,dx00);
 822             ty               = _mm_mul_ps(fscal,dy00);
 823             tz               = _mm_mul_ps(fscal,dz00);
 824
 825             /* Update vectorial force */
 826             fix0             = _mm_add_ps(fix0,tx);
 827             fiy0             = _mm_add_ps(fiy0,ty);
 828             fiz0             = _mm_add_ps(fiz0,tz);
 829
 830             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 831             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 832             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 833             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 834             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 835
 836             }
 837
 838             /* Inner loop uses 63 flops */
 839         }
 840
 841         /* End of innermost loop */
 842
 843         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 844                                               f+i_coord_offset,fshift+i_shift_offset);
 845
 846         /* Increment number of inner iterations */
 847         inneriter                  += j_index_end - j_index_start;
 848
 849         /* Outer loop uses 7 flops */
 850     }
 851
 852     /* Increment number of outer iterations */
 853     outeriter        += nri;
 854
 855     /* Update outer/inner flops */
 856
 857     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*63);
 858 }