src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_single.h"
  48 #include "kernelutil_x86_sse4_1_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_VF_sse4_1_single
  52  * Electrostatics interaction: CubicSplineTable
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_VF_sse4_1_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwioffset3;
  90     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  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           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  97     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     int              nvdwtype;
 100     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 101     int              *vdwtype;
 102     real             *vdwparam;
 103     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 104     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 105     __m128i          vfitab;
 106     __m128i          ifour       = _mm_set1_epi32(4);
 107     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 108     real             *vftab;
 109     __m128           dummy_mask,cutoff_mask;
 110     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 111     __m128           one     = _mm_set1_ps(1.0);
 112     __m128           two     = _mm_set1_ps(2.0);
 113     x                = xx[0];
 114     f                = ff[0];
 115
 116     nri              = nlist->nri;
 117     iinr             = nlist->iinr;
 118     jindex           = nlist->jindex;
 119     jjnr             = nlist->jjnr;
 120     shiftidx         = nlist->shift;
 121     gid              = nlist->gid;
 122     shiftvec         = fr->shift_vec[0];
 123     fshift           = fr->fshift[0];
 124     facel            = _mm_set1_ps(fr->epsfac);
 125     charge           = mdatoms->chargeA;
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129
 130     vftab            = kernel_data->table_elec->data;
 131     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 132
 133     /* Setup water-specific parameters */
 134     inr              = nlist->iinr[0];
 135     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 136     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 137     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 138     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 139
 140     /* Avoid stupid compiler warnings */
 141     jnrA = jnrB = jnrC = jnrD = 0;
 142     j_coord_offsetA = 0;
 143     j_coord_offsetB = 0;
 144     j_coord_offsetC = 0;
 145     j_coord_offsetD = 0;
 146
 147     outeriter        = 0;
 148     inneriter        = 0;
 149
 150     for(iidx=0;iidx<4*DIM;iidx++)
 151     {
 152         scratch[iidx] = 0.0;
 153     }
 154
 155     /* Start outer loop over neighborlists */
 156     for(iidx=0; iidx<nri; iidx++)
 157     {
 158         /* Load shift vector for this list */
 159         i_shift_offset   = DIM*shiftidx[iidx];
 160
 161         /* Load limits for loop over neighbors */
 162         j_index_start    = jindex[iidx];
 163         j_index_end      = jindex[iidx+1];
 164
 165         /* Get outer coordinate index */
 166         inr              = iinr[iidx];
 167         i_coord_offset   = DIM*inr;
 168
 169         /* Load i particle coords and add shift vector */
 170         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 171                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 172
 173         fix0             = _mm_setzero_ps();
 174         fiy0             = _mm_setzero_ps();
 175         fiz0             = _mm_setzero_ps();
 176         fix1             = _mm_setzero_ps();
 177         fiy1             = _mm_setzero_ps();
 178         fiz1             = _mm_setzero_ps();
 179         fix2             = _mm_setzero_ps();
 180         fiy2             = _mm_setzero_ps();
 181         fiz2             = _mm_setzero_ps();
 182         fix3             = _mm_setzero_ps();
 183         fiy3             = _mm_setzero_ps();
 184         fiz3             = _mm_setzero_ps();
 185
 186         /* Reset potential sums */
 187         velecsum         = _mm_setzero_ps();
 188         vvdwsum          = _mm_setzero_ps();
 189
 190         /* Start inner kernel loop */
 191         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 192         {
 193
 194             /* Get j neighbor index, and coordinate index */
 195             jnrA             = jjnr[jidx];
 196             jnrB             = jjnr[jidx+1];
 197             jnrC             = jjnr[jidx+2];
 198             jnrD             = jjnr[jidx+3];
 199             j_coord_offsetA  = DIM*jnrA;
 200             j_coord_offsetB  = DIM*jnrB;
 201             j_coord_offsetC  = DIM*jnrC;
 202             j_coord_offsetD  = DIM*jnrD;
 203
 204             /* load j atom coordinates */
 205             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 206                                               x+j_coord_offsetC,x+j_coord_offsetD,
 207                                               &jx0,&jy0,&jz0);
 208
 209             /* Calculate displacement vector */
 210             dx00             = _mm_sub_ps(ix0,jx0);
 211             dy00             = _mm_sub_ps(iy0,jy0);
 212             dz00             = _mm_sub_ps(iz0,jz0);
 213             dx10             = _mm_sub_ps(ix1,jx0);
 214             dy10             = _mm_sub_ps(iy1,jy0);
 215             dz10             = _mm_sub_ps(iz1,jz0);
 216             dx20             = _mm_sub_ps(ix2,jx0);
 217             dy20             = _mm_sub_ps(iy2,jy0);
 218             dz20             = _mm_sub_ps(iz2,jz0);
 219             dx30             = _mm_sub_ps(ix3,jx0);
 220             dy30             = _mm_sub_ps(iy3,jy0);
 221             dz30             = _mm_sub_ps(iz3,jz0);
 222
 223             /* Calculate squared distance and things based on it */
 224             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 225             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 226             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 227             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 228
 229             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 230             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 231             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 232
 233             rinvsq00         = gmx_mm_inv_ps(rsq00);
 234
 235             /* Load parameters for j particles */
 236             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 237                                                               charge+jnrC+0,charge+jnrD+0);
 238             vdwjidx0A        = 2*vdwtype[jnrA+0];
 239             vdwjidx0B        = 2*vdwtype[jnrB+0];
 240             vdwjidx0C        = 2*vdwtype[jnrC+0];
 241             vdwjidx0D        = 2*vdwtype[jnrD+0];
 242
 243             fjx0             = _mm_setzero_ps();
 244             fjy0             = _mm_setzero_ps();
 245             fjz0             = _mm_setzero_ps();
 246
 247             /**************************
 248              * CALCULATE INTERACTIONS *
 249              **************************/
 250
 251             /* Compute parameters for interactions between i and j atoms */
 252             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 253                                          vdwparam+vdwioffset0+vdwjidx0B,
 254                                          vdwparam+vdwioffset0+vdwjidx0C,
 255                                          vdwparam+vdwioffset0+vdwjidx0D,
 256                                          &c6_00,&c12_00);
 257
 258             /* LENNARD-JONES DISPERSION/REPULSION */
 259
 260             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 261             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 262             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 263             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 264             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 265
 266             /* Update potential sum for this i atom from the interaction with this j atom. */
 267             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 268
 269             fscal            = fvdw;
 270
 271             /* Calculate temporary vectorial force */
 272             tx               = _mm_mul_ps(fscal,dx00);
 273             ty               = _mm_mul_ps(fscal,dy00);
 274             tz               = _mm_mul_ps(fscal,dz00);
 275
 276             /* Update vectorial force */
 277             fix0             = _mm_add_ps(fix0,tx);
 278             fiy0             = _mm_add_ps(fiy0,ty);
 279             fiz0             = _mm_add_ps(fiz0,tz);
 280
 281             fjx0             = _mm_add_ps(fjx0,tx);
 282             fjy0             = _mm_add_ps(fjy0,ty);
 283             fjz0             = _mm_add_ps(fjz0,tz);
 284
 285             /**************************
 286              * CALCULATE INTERACTIONS *
 287              **************************/
 288
 289             r10              = _mm_mul_ps(rsq10,rinv10);
 290
 291             /* Compute parameters for interactions between i and j atoms */
 292             qq10             = _mm_mul_ps(iq1,jq0);
 293
 294             /* Calculate table index by multiplying r with table scale and truncate to integer */
 295             rt               = _mm_mul_ps(r10,vftabscale);
 296             vfitab           = _mm_cvttps_epi32(rt);
 297             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 298             vfitab           = _mm_slli_epi32(vfitab,2);
 299
 300             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 301             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 302             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 303             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 304             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 305             _MM_TRANSPOSE4_PS(Y,F,G,H);
 306             Heps             = _mm_mul_ps(vfeps,H);
 307             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 308             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 309             velec            = _mm_mul_ps(qq10,VV);
 310             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 311             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 312
 313             /* Update potential sum for this i atom from the interaction with this j atom. */
 314             velecsum         = _mm_add_ps(velecsum,velec);
 315
 316             fscal            = felec;
 317
 318             /* Calculate temporary vectorial force */
 319             tx               = _mm_mul_ps(fscal,dx10);
 320             ty               = _mm_mul_ps(fscal,dy10);
 321             tz               = _mm_mul_ps(fscal,dz10);
 322
 323             /* Update vectorial force */
 324             fix1             = _mm_add_ps(fix1,tx);
 325             fiy1             = _mm_add_ps(fiy1,ty);
 326             fiz1             = _mm_add_ps(fiz1,tz);
 327
 328             fjx0             = _mm_add_ps(fjx0,tx);
 329             fjy0             = _mm_add_ps(fjy0,ty);
 330             fjz0             = _mm_add_ps(fjz0,tz);
 331
 332             /**************************
 333              * CALCULATE INTERACTIONS *
 334              **************************/
 335
 336             r20              = _mm_mul_ps(rsq20,rinv20);
 337
 338             /* Compute parameters for interactions between i and j atoms */
 339             qq20             = _mm_mul_ps(iq2,jq0);
 340
 341             /* Calculate table index by multiplying r with table scale and truncate to integer */
 342             rt               = _mm_mul_ps(r20,vftabscale);
 343             vfitab           = _mm_cvttps_epi32(rt);
 344             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 345             vfitab           = _mm_slli_epi32(vfitab,2);
 346
 347             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 348             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 349             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 350             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 351             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 352             _MM_TRANSPOSE4_PS(Y,F,G,H);
 353             Heps             = _mm_mul_ps(vfeps,H);
 354             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 355             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 356             velec            = _mm_mul_ps(qq20,VV);
 357             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 358             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 359
 360             /* Update potential sum for this i atom from the interaction with this j atom. */
 361             velecsum         = _mm_add_ps(velecsum,velec);
 362
 363             fscal            = felec;
 364
 365             /* Calculate temporary vectorial force */
 366             tx               = _mm_mul_ps(fscal,dx20);
 367             ty               = _mm_mul_ps(fscal,dy20);
 368             tz               = _mm_mul_ps(fscal,dz20);
 369
 370             /* Update vectorial force */
 371             fix2             = _mm_add_ps(fix2,tx);
 372             fiy2             = _mm_add_ps(fiy2,ty);
 373             fiz2             = _mm_add_ps(fiz2,tz);
 374
 375             fjx0             = _mm_add_ps(fjx0,tx);
 376             fjy0             = _mm_add_ps(fjy0,ty);
 377             fjz0             = _mm_add_ps(fjz0,tz);
 378
 379             /**************************
 380              * CALCULATE INTERACTIONS *
 381              **************************/
 382
 383             r30              = _mm_mul_ps(rsq30,rinv30);
 384
 385             /* Compute parameters for interactions between i and j atoms */
 386             qq30             = _mm_mul_ps(iq3,jq0);
 387
 388             /* Calculate table index by multiplying r with table scale and truncate to integer */
 389             rt               = _mm_mul_ps(r30,vftabscale);
 390             vfitab           = _mm_cvttps_epi32(rt);
 391             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 392             vfitab           = _mm_slli_epi32(vfitab,2);
 393
 394             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 395             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 396             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 397             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 398             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 399             _MM_TRANSPOSE4_PS(Y,F,G,H);
 400             Heps             = _mm_mul_ps(vfeps,H);
 401             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 402             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 403             velec            = _mm_mul_ps(qq30,VV);
 404             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 405             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 406
 407             /* Update potential sum for this i atom from the interaction with this j atom. */
 408             velecsum         = _mm_add_ps(velecsum,velec);
 409
 410             fscal            = felec;
 411
 412             /* Calculate temporary vectorial force */
 413             tx               = _mm_mul_ps(fscal,dx30);
 414             ty               = _mm_mul_ps(fscal,dy30);
 415             tz               = _mm_mul_ps(fscal,dz30);
 416
 417             /* Update vectorial force */
 418             fix3             = _mm_add_ps(fix3,tx);
 419             fiy3             = _mm_add_ps(fiy3,ty);
 420             fiz3             = _mm_add_ps(fiz3,tz);
 421
 422             fjx0             = _mm_add_ps(fjx0,tx);
 423             fjy0             = _mm_add_ps(fjy0,ty);
 424             fjz0             = _mm_add_ps(fjz0,tz);
 425
 426             fjptrA             = f+j_coord_offsetA;
 427             fjptrB             = f+j_coord_offsetB;
 428             fjptrC             = f+j_coord_offsetC;
 429             fjptrD             = f+j_coord_offsetD;
 430
 431             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 432
 433             /* Inner loop uses 161 flops */
 434         }
 435
 436         if(jidx<j_index_end)
 437         {
 438
 439             /* Get j neighbor index, and coordinate index */
 440             jnrlistA         = jjnr[jidx];
 441             jnrlistB         = jjnr[jidx+1];
 442             jnrlistC         = jjnr[jidx+2];
 443             jnrlistD         = jjnr[jidx+3];
 444             /* Sign of each element will be negative for non-real atoms.
 445              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 446              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 447              */
 448             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 449             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 450             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 451             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 452             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 453             j_coord_offsetA  = DIM*jnrA;
 454             j_coord_offsetB  = DIM*jnrB;
 455             j_coord_offsetC  = DIM*jnrC;
 456             j_coord_offsetD  = DIM*jnrD;
 457
 458             /* load j atom coordinates */
 459             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 460                                               x+j_coord_offsetC,x+j_coord_offsetD,
 461                                               &jx0,&jy0,&jz0);
 462
 463             /* Calculate displacement vector */
 464             dx00             = _mm_sub_ps(ix0,jx0);
 465             dy00             = _mm_sub_ps(iy0,jy0);
 466             dz00             = _mm_sub_ps(iz0,jz0);
 467             dx10             = _mm_sub_ps(ix1,jx0);
 468             dy10             = _mm_sub_ps(iy1,jy0);
 469             dz10             = _mm_sub_ps(iz1,jz0);
 470             dx20             = _mm_sub_ps(ix2,jx0);
 471             dy20             = _mm_sub_ps(iy2,jy0);
 472             dz20             = _mm_sub_ps(iz2,jz0);
 473             dx30             = _mm_sub_ps(ix3,jx0);
 474             dy30             = _mm_sub_ps(iy3,jy0);
 475             dz30             = _mm_sub_ps(iz3,jz0);
 476
 477             /* Calculate squared distance and things based on it */
 478             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 479             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 480             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 481             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 482
 483             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 484             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 485             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 486
 487             rinvsq00         = gmx_mm_inv_ps(rsq00);
 488
 489             /* Load parameters for j particles */
 490             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 491                                                               charge+jnrC+0,charge+jnrD+0);
 492             vdwjidx0A        = 2*vdwtype[jnrA+0];
 493             vdwjidx0B        = 2*vdwtype[jnrB+0];
 494             vdwjidx0C        = 2*vdwtype[jnrC+0];
 495             vdwjidx0D        = 2*vdwtype[jnrD+0];
 496
 497             fjx0             = _mm_setzero_ps();
 498             fjy0             = _mm_setzero_ps();
 499             fjz0             = _mm_setzero_ps();
 500
 501             /**************************
 502              * CALCULATE INTERACTIONS *
 503              **************************/
 504
 505             /* Compute parameters for interactions between i and j atoms */
 506             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 507                                          vdwparam+vdwioffset0+vdwjidx0B,
 508                                          vdwparam+vdwioffset0+vdwjidx0C,
 509                                          vdwparam+vdwioffset0+vdwjidx0D,
 510                                          &c6_00,&c12_00);
 511
 512             /* LENNARD-JONES DISPERSION/REPULSION */
 513
 514             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 515             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 516             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 517             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 518             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 519
 520             /* Update potential sum for this i atom from the interaction with this j atom. */
 521             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 522             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 523
 524             fscal            = fvdw;
 525
 526             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 527
 528             /* Calculate temporary vectorial force */
 529             tx               = _mm_mul_ps(fscal,dx00);
 530             ty               = _mm_mul_ps(fscal,dy00);
 531             tz               = _mm_mul_ps(fscal,dz00);
 532
 533             /* Update vectorial force */
 534             fix0             = _mm_add_ps(fix0,tx);
 535             fiy0             = _mm_add_ps(fiy0,ty);
 536             fiz0             = _mm_add_ps(fiz0,tz);
 537
 538             fjx0             = _mm_add_ps(fjx0,tx);
 539             fjy0             = _mm_add_ps(fjy0,ty);
 540             fjz0             = _mm_add_ps(fjz0,tz);
 541
 542             /**************************
 543              * CALCULATE INTERACTIONS *
 544              **************************/
 545
 546             r10              = _mm_mul_ps(rsq10,rinv10);
 547             r10              = _mm_andnot_ps(dummy_mask,r10);
 548
 549             /* Compute parameters for interactions between i and j atoms */
 550             qq10             = _mm_mul_ps(iq1,jq0);
 551
 552             /* Calculate table index by multiplying r with table scale and truncate to integer */
 553             rt               = _mm_mul_ps(r10,vftabscale);
 554             vfitab           = _mm_cvttps_epi32(rt);
 555             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 556             vfitab           = _mm_slli_epi32(vfitab,2);
 557
 558             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 559             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 560             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 561             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 562             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 563             _MM_TRANSPOSE4_PS(Y,F,G,H);
 564             Heps             = _mm_mul_ps(vfeps,H);
 565             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 566             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 567             velec            = _mm_mul_ps(qq10,VV);
 568             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 569             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 570
 571             /* Update potential sum for this i atom from the interaction with this j atom. */
 572             velec            = _mm_andnot_ps(dummy_mask,velec);
 573             velecsum         = _mm_add_ps(velecsum,velec);
 574
 575             fscal            = felec;
 576
 577             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 578
 579             /* Calculate temporary vectorial force */
 580             tx               = _mm_mul_ps(fscal,dx10);
 581             ty               = _mm_mul_ps(fscal,dy10);
 582             tz               = _mm_mul_ps(fscal,dz10);
 583
 584             /* Update vectorial force */
 585             fix1             = _mm_add_ps(fix1,tx);
 586             fiy1             = _mm_add_ps(fiy1,ty);
 587             fiz1             = _mm_add_ps(fiz1,tz);
 588
 589             fjx0             = _mm_add_ps(fjx0,tx);
 590             fjy0             = _mm_add_ps(fjy0,ty);
 591             fjz0             = _mm_add_ps(fjz0,tz);
 592
 593             /**************************
 594              * CALCULATE INTERACTIONS *
 595              **************************/
 596
 597             r20              = _mm_mul_ps(rsq20,rinv20);
 598             r20              = _mm_andnot_ps(dummy_mask,r20);
 599
 600             /* Compute parameters for interactions between i and j atoms */
 601             qq20             = _mm_mul_ps(iq2,jq0);
 602
 603             /* Calculate table index by multiplying r with table scale and truncate to integer */
 604             rt               = _mm_mul_ps(r20,vftabscale);
 605             vfitab           = _mm_cvttps_epi32(rt);
 606             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 607             vfitab           = _mm_slli_epi32(vfitab,2);
 608
 609             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 610             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 611             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 612             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 613             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 614             _MM_TRANSPOSE4_PS(Y,F,G,H);
 615             Heps             = _mm_mul_ps(vfeps,H);
 616             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 617             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 618             velec            = _mm_mul_ps(qq20,VV);
 619             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 620             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 621
 622             /* Update potential sum for this i atom from the interaction with this j atom. */
 623             velec            = _mm_andnot_ps(dummy_mask,velec);
 624             velecsum         = _mm_add_ps(velecsum,velec);
 625
 626             fscal            = felec;
 627
 628             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 629
 630             /* Calculate temporary vectorial force */
 631             tx               = _mm_mul_ps(fscal,dx20);
 632             ty               = _mm_mul_ps(fscal,dy20);
 633             tz               = _mm_mul_ps(fscal,dz20);
 634
 635             /* Update vectorial force */
 636             fix2             = _mm_add_ps(fix2,tx);
 637             fiy2             = _mm_add_ps(fiy2,ty);
 638             fiz2             = _mm_add_ps(fiz2,tz);
 639
 640             fjx0             = _mm_add_ps(fjx0,tx);
 641             fjy0             = _mm_add_ps(fjy0,ty);
 642             fjz0             = _mm_add_ps(fjz0,tz);
 643
 644             /**************************
 645              * CALCULATE INTERACTIONS *
 646              **************************/
 647
 648             r30              = _mm_mul_ps(rsq30,rinv30);
 649             r30              = _mm_andnot_ps(dummy_mask,r30);
 650
 651             /* Compute parameters for interactions between i and j atoms */
 652             qq30             = _mm_mul_ps(iq3,jq0);
 653
 654             /* Calculate table index by multiplying r with table scale and truncate to integer */
 655             rt               = _mm_mul_ps(r30,vftabscale);
 656             vfitab           = _mm_cvttps_epi32(rt);
 657             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 658             vfitab           = _mm_slli_epi32(vfitab,2);
 659
 660             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 661             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 662             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 663             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 664             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 665             _MM_TRANSPOSE4_PS(Y,F,G,H);
 666             Heps             = _mm_mul_ps(vfeps,H);
 667             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 668             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 669             velec            = _mm_mul_ps(qq30,VV);
 670             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 671             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 672
 673             /* Update potential sum for this i atom from the interaction with this j atom. */
 674             velec            = _mm_andnot_ps(dummy_mask,velec);
 675             velecsum         = _mm_add_ps(velecsum,velec);
 676
 677             fscal            = felec;
 678
 679             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 680
 681             /* Calculate temporary vectorial force */
 682             tx               = _mm_mul_ps(fscal,dx30);
 683             ty               = _mm_mul_ps(fscal,dy30);
 684             tz               = _mm_mul_ps(fscal,dz30);
 685
 686             /* Update vectorial force */
 687             fix3             = _mm_add_ps(fix3,tx);
 688             fiy3             = _mm_add_ps(fiy3,ty);
 689             fiz3             = _mm_add_ps(fiz3,tz);
 690
 691             fjx0             = _mm_add_ps(fjx0,tx);
 692             fjy0             = _mm_add_ps(fjy0,ty);
 693             fjz0             = _mm_add_ps(fjz0,tz);
 694
 695             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 696             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 697             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 698             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 699
 700             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 701
 702             /* Inner loop uses 164 flops */
 703         }
 704
 705         /* End of innermost loop */
 706
 707         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 708                                               f+i_coord_offset,fshift+i_shift_offset);
 709
 710         ggid                        = gid[iidx];
 711         /* Update potential energies */
 712         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 713         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 714
 715         /* Increment number of inner iterations */
 716         inneriter                  += j_index_end - j_index_start;
 717
 718         /* Outer loop uses 26 flops */
 719     }
 720
 721     /* Increment number of outer iterations */
 722     outeriter        += nri;
 723
 724     /* Update outer/inner flops */
 725
 726     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*164);
 727 }
 728 /*
 729  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_sse4_1_single
 730  * Electrostatics interaction: CubicSplineTable
 731  * VdW interaction:            LennardJones
 732  * Geometry:                   Water4-Particle
 733  * Calculate force/pot:        Force
 734  */
 735 void
 736 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_sse4_1_single
 737                     (t_nblist                    * gmx_restrict       nlist,
 738                      rvec                        * gmx_restrict          xx,
 739                      rvec                        * gmx_restrict          ff,
 740                      t_forcerec                  * gmx_restrict          fr,
 741                      t_mdatoms                   * gmx_restrict     mdatoms,
 742                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 743                      t_nrnb                      * gmx_restrict        nrnb)
 744 {
 745     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 746      * just 0 for non-waters.
 747      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 748      * jnr indices corresponding to data put in the four positions in the SIMD register.
 749      */
 750     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 751     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 752     int              jnrA,jnrB,jnrC,jnrD;
 753     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 754     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 755     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 756     real             rcutoff_scalar;
 757     real             *shiftvec,*fshift,*x,*f;
 758     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 759     real             scratch[4*DIM];
 760     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 761     int              vdwioffset0;
 762     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 763     int              vdwioffset1;
 764     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 765     int              vdwioffset2;
 766     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 767     int              vdwioffset3;
 768     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 769     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 770     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 771     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 772     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 773     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 774     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 775     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 776     real             *charge;
 777     int              nvdwtype;
 778     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 779     int              *vdwtype;
 780     real             *vdwparam;
 781     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 782     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 783     __m128i          vfitab;
 784     __m128i          ifour       = _mm_set1_epi32(4);
 785     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 786     real             *vftab;
 787     __m128           dummy_mask,cutoff_mask;
 788     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 789     __m128           one     = _mm_set1_ps(1.0);
 790     __m128           two     = _mm_set1_ps(2.0);
 791     x                = xx[0];
 792     f                = ff[0];
 793
 794     nri              = nlist->nri;
 795     iinr             = nlist->iinr;
 796     jindex           = nlist->jindex;
 797     jjnr             = nlist->jjnr;
 798     shiftidx         = nlist->shift;
 799     gid              = nlist->gid;
 800     shiftvec         = fr->shift_vec[0];
 801     fshift           = fr->fshift[0];
 802     facel            = _mm_set1_ps(fr->epsfac);
 803     charge           = mdatoms->chargeA;
 804     nvdwtype         = fr->ntype;
 805     vdwparam         = fr->nbfp;
 806     vdwtype          = mdatoms->typeA;
 807
 808     vftab            = kernel_data->table_elec->data;
 809     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 810
 811     /* Setup water-specific parameters */
 812     inr              = nlist->iinr[0];
 813     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 814     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 815     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 816     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 817
 818     /* Avoid stupid compiler warnings */
 819     jnrA = jnrB = jnrC = jnrD = 0;
 820     j_coord_offsetA = 0;
 821     j_coord_offsetB = 0;
 822     j_coord_offsetC = 0;
 823     j_coord_offsetD = 0;
 824
 825     outeriter        = 0;
 826     inneriter        = 0;
 827
 828     for(iidx=0;iidx<4*DIM;iidx++)
 829     {
 830         scratch[iidx] = 0.0;
 831     }
 832
 833     /* Start outer loop over neighborlists */
 834     for(iidx=0; iidx<nri; iidx++)
 835     {
 836         /* Load shift vector for this list */
 837         i_shift_offset   = DIM*shiftidx[iidx];
 838
 839         /* Load limits for loop over neighbors */
 840         j_index_start    = jindex[iidx];
 841         j_index_end      = jindex[iidx+1];
 842
 843         /* Get outer coordinate index */
 844         inr              = iinr[iidx];
 845         i_coord_offset   = DIM*inr;
 846
 847         /* Load i particle coords and add shift vector */
 848         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 849                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 850
 851         fix0             = _mm_setzero_ps();
 852         fiy0             = _mm_setzero_ps();
 853         fiz0             = _mm_setzero_ps();
 854         fix1             = _mm_setzero_ps();
 855         fiy1             = _mm_setzero_ps();
 856         fiz1             = _mm_setzero_ps();
 857         fix2             = _mm_setzero_ps();
 858         fiy2             = _mm_setzero_ps();
 859         fiz2             = _mm_setzero_ps();
 860         fix3             = _mm_setzero_ps();
 861         fiy3             = _mm_setzero_ps();
 862         fiz3             = _mm_setzero_ps();
 863
 864         /* Start inner kernel loop */
 865         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 866         {
 867
 868             /* Get j neighbor index, and coordinate index */
 869             jnrA             = jjnr[jidx];
 870             jnrB             = jjnr[jidx+1];
 871             jnrC             = jjnr[jidx+2];
 872             jnrD             = jjnr[jidx+3];
 873             j_coord_offsetA  = DIM*jnrA;
 874             j_coord_offsetB  = DIM*jnrB;
 875             j_coord_offsetC  = DIM*jnrC;
 876             j_coord_offsetD  = DIM*jnrD;
 877
 878             /* load j atom coordinates */
 879             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 880                                               x+j_coord_offsetC,x+j_coord_offsetD,
 881                                               &jx0,&jy0,&jz0);
 882
 883             /* Calculate displacement vector */
 884             dx00             = _mm_sub_ps(ix0,jx0);
 885             dy00             = _mm_sub_ps(iy0,jy0);
 886             dz00             = _mm_sub_ps(iz0,jz0);
 887             dx10             = _mm_sub_ps(ix1,jx0);
 888             dy10             = _mm_sub_ps(iy1,jy0);
 889             dz10             = _mm_sub_ps(iz1,jz0);
 890             dx20             = _mm_sub_ps(ix2,jx0);
 891             dy20             = _mm_sub_ps(iy2,jy0);
 892             dz20             = _mm_sub_ps(iz2,jz0);
 893             dx30             = _mm_sub_ps(ix3,jx0);
 894             dy30             = _mm_sub_ps(iy3,jy0);
 895             dz30             = _mm_sub_ps(iz3,jz0);
 896
 897             /* Calculate squared distance and things based on it */
 898             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 899             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 900             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 901             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 902
 903             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 904             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 905             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 906
 907             rinvsq00         = gmx_mm_inv_ps(rsq00);
 908
 909             /* Load parameters for j particles */
 910             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 911                                                               charge+jnrC+0,charge+jnrD+0);
 912             vdwjidx0A        = 2*vdwtype[jnrA+0];
 913             vdwjidx0B        = 2*vdwtype[jnrB+0];
 914             vdwjidx0C        = 2*vdwtype[jnrC+0];
 915             vdwjidx0D        = 2*vdwtype[jnrD+0];
 916
 917             fjx0             = _mm_setzero_ps();
 918             fjy0             = _mm_setzero_ps();
 919             fjz0             = _mm_setzero_ps();
 920
 921             /**************************
 922              * CALCULATE INTERACTIONS *
 923              **************************/
 924
 925             /* Compute parameters for interactions between i and j atoms */
 926             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 927                                          vdwparam+vdwioffset0+vdwjidx0B,
 928                                          vdwparam+vdwioffset0+vdwjidx0C,
 929                                          vdwparam+vdwioffset0+vdwjidx0D,
 930                                          &c6_00,&c12_00);
 931
 932             /* LENNARD-JONES DISPERSION/REPULSION */
 933
 934             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 935             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 936
 937             fscal            = fvdw;
 938
 939             /* Calculate temporary vectorial force */
 940             tx               = _mm_mul_ps(fscal,dx00);
 941             ty               = _mm_mul_ps(fscal,dy00);
 942             tz               = _mm_mul_ps(fscal,dz00);
 943
 944             /* Update vectorial force */
 945             fix0             = _mm_add_ps(fix0,tx);
 946             fiy0             = _mm_add_ps(fiy0,ty);
 947             fiz0             = _mm_add_ps(fiz0,tz);
 948
 949             fjx0             = _mm_add_ps(fjx0,tx);
 950             fjy0             = _mm_add_ps(fjy0,ty);
 951             fjz0             = _mm_add_ps(fjz0,tz);
 952
 953             /**************************
 954              * CALCULATE INTERACTIONS *
 955              **************************/
 956
 957             r10              = _mm_mul_ps(rsq10,rinv10);
 958
 959             /* Compute parameters for interactions between i and j atoms */
 960             qq10             = _mm_mul_ps(iq1,jq0);
 961
 962             /* Calculate table index by multiplying r with table scale and truncate to integer */
 963             rt               = _mm_mul_ps(r10,vftabscale);
 964             vfitab           = _mm_cvttps_epi32(rt);
 965             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 966             vfitab           = _mm_slli_epi32(vfitab,2);
 967
 968             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 969             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 970             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 971             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 972             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 973             _MM_TRANSPOSE4_PS(Y,F,G,H);
 974             Heps             = _mm_mul_ps(vfeps,H);
 975             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 976             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 977             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 978
 979             fscal            = felec;
 980
 981             /* Calculate temporary vectorial force */
 982             tx               = _mm_mul_ps(fscal,dx10);
 983             ty               = _mm_mul_ps(fscal,dy10);
 984             tz               = _mm_mul_ps(fscal,dz10);
 985
 986             /* Update vectorial force */
 987             fix1             = _mm_add_ps(fix1,tx);
 988             fiy1             = _mm_add_ps(fiy1,ty);
 989             fiz1             = _mm_add_ps(fiz1,tz);
 990
 991             fjx0             = _mm_add_ps(fjx0,tx);
 992             fjy0             = _mm_add_ps(fjy0,ty);
 993             fjz0             = _mm_add_ps(fjz0,tz);
 994
 995             /**************************
 996              * CALCULATE INTERACTIONS *
 997              **************************/
 998
 999             r20              = _mm_mul_ps(rsq20,rinv20);
1000
1001             /* Compute parameters for interactions between i and j atoms */
1002             qq20             = _mm_mul_ps(iq2,jq0);
1003
1004             /* Calculate table index by multiplying r with table scale and truncate to integer */
1005             rt               = _mm_mul_ps(r20,vftabscale);
1006             vfitab           = _mm_cvttps_epi32(rt);
1007             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1008             vfitab           = _mm_slli_epi32(vfitab,2);
1009
1010             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1011             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1012             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1013             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1014             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1015             _MM_TRANSPOSE4_PS(Y,F,G,H);
1016             Heps             = _mm_mul_ps(vfeps,H);
1017             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1018             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1019             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1020
1021             fscal            = felec;
1022
1023             /* Calculate temporary vectorial force */
1024             tx               = _mm_mul_ps(fscal,dx20);
1025             ty               = _mm_mul_ps(fscal,dy20);
1026             tz               = _mm_mul_ps(fscal,dz20);
1027
1028             /* Update vectorial force */
1029             fix2             = _mm_add_ps(fix2,tx);
1030             fiy2             = _mm_add_ps(fiy2,ty);
1031             fiz2             = _mm_add_ps(fiz2,tz);
1032
1033             fjx0             = _mm_add_ps(fjx0,tx);
1034             fjy0             = _mm_add_ps(fjy0,ty);
1035             fjz0             = _mm_add_ps(fjz0,tz);
1036
1037             /**************************
1038              * CALCULATE INTERACTIONS *
1039              **************************/
1040
1041             r30              = _mm_mul_ps(rsq30,rinv30);
1042
1043             /* Compute parameters for interactions between i and j atoms */
1044             qq30             = _mm_mul_ps(iq3,jq0);
1045
1046             /* Calculate table index by multiplying r with table scale and truncate to integer */
1047             rt               = _mm_mul_ps(r30,vftabscale);
1048             vfitab           = _mm_cvttps_epi32(rt);
1049             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1050             vfitab           = _mm_slli_epi32(vfitab,2);
1051
1052             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1053             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1054             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1055             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1056             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1057             _MM_TRANSPOSE4_PS(Y,F,G,H);
1058             Heps             = _mm_mul_ps(vfeps,H);
1059             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1060             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1061             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1062
1063             fscal            = felec;
1064
1065             /* Calculate temporary vectorial force */
1066             tx               = _mm_mul_ps(fscal,dx30);
1067             ty               = _mm_mul_ps(fscal,dy30);
1068             tz               = _mm_mul_ps(fscal,dz30);
1069
1070             /* Update vectorial force */
1071             fix3             = _mm_add_ps(fix3,tx);
1072             fiy3             = _mm_add_ps(fiy3,ty);
1073             fiz3             = _mm_add_ps(fiz3,tz);
1074
1075             fjx0             = _mm_add_ps(fjx0,tx);
1076             fjy0             = _mm_add_ps(fjy0,ty);
1077             fjz0             = _mm_add_ps(fjz0,tz);
1078
1079             fjptrA             = f+j_coord_offsetA;
1080             fjptrB             = f+j_coord_offsetB;
1081             fjptrC             = f+j_coord_offsetC;
1082             fjptrD             = f+j_coord_offsetD;
1083
1084             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1085
1086             /* Inner loop uses 144 flops */
1087         }
1088
1089         if(jidx<j_index_end)
1090         {
1091
1092             /* Get j neighbor index, and coordinate index */
1093             jnrlistA         = jjnr[jidx];
1094             jnrlistB         = jjnr[jidx+1];
1095             jnrlistC         = jjnr[jidx+2];
1096             jnrlistD         = jjnr[jidx+3];
1097             /* Sign of each element will be negative for non-real atoms.
1098              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1099              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1100              */
1101             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1102             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1103             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1104             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1105             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1106             j_coord_offsetA  = DIM*jnrA;
1107             j_coord_offsetB  = DIM*jnrB;
1108             j_coord_offsetC  = DIM*jnrC;
1109             j_coord_offsetD  = DIM*jnrD;
1110
1111             /* load j atom coordinates */
1112             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1113                                               x+j_coord_offsetC,x+j_coord_offsetD,
1114                                               &jx0,&jy0,&jz0);
1115
1116             /* Calculate displacement vector */
1117             dx00             = _mm_sub_ps(ix0,jx0);
1118             dy00             = _mm_sub_ps(iy0,jy0);
1119             dz00             = _mm_sub_ps(iz0,jz0);
1120             dx10             = _mm_sub_ps(ix1,jx0);
1121             dy10             = _mm_sub_ps(iy1,jy0);
1122             dz10             = _mm_sub_ps(iz1,jz0);
1123             dx20             = _mm_sub_ps(ix2,jx0);
1124             dy20             = _mm_sub_ps(iy2,jy0);
1125             dz20             = _mm_sub_ps(iz2,jz0);
1126             dx30             = _mm_sub_ps(ix3,jx0);
1127             dy30             = _mm_sub_ps(iy3,jy0);
1128             dz30             = _mm_sub_ps(iz3,jz0);
1129
1130             /* Calculate squared distance and things based on it */
1131             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1132             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1133             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1134             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1135
1136             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1137             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1138             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1139
1140             rinvsq00         = gmx_mm_inv_ps(rsq00);
1141
1142             /* Load parameters for j particles */
1143             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1144                                                               charge+jnrC+0,charge+jnrD+0);
1145             vdwjidx0A        = 2*vdwtype[jnrA+0];
1146             vdwjidx0B        = 2*vdwtype[jnrB+0];
1147             vdwjidx0C        = 2*vdwtype[jnrC+0];
1148             vdwjidx0D        = 2*vdwtype[jnrD+0];
1149
1150             fjx0             = _mm_setzero_ps();
1151             fjy0             = _mm_setzero_ps();
1152             fjz0             = _mm_setzero_ps();
1153
1154             /**************************
1155              * CALCULATE INTERACTIONS *
1156              **************************/
1157
1158             /* Compute parameters for interactions between i and j atoms */
1159             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1160                                          vdwparam+vdwioffset0+vdwjidx0B,
1161                                          vdwparam+vdwioffset0+vdwjidx0C,
1162                                          vdwparam+vdwioffset0+vdwjidx0D,
1163                                          &c6_00,&c12_00);
1164
1165             /* LENNARD-JONES DISPERSION/REPULSION */
1166
1167             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1168             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1169
1170             fscal            = fvdw;
1171
1172             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1173
1174             /* Calculate temporary vectorial force */
1175             tx               = _mm_mul_ps(fscal,dx00);
1176             ty               = _mm_mul_ps(fscal,dy00);
1177             tz               = _mm_mul_ps(fscal,dz00);
1178
1179             /* Update vectorial force */
1180             fix0             = _mm_add_ps(fix0,tx);
1181             fiy0             = _mm_add_ps(fiy0,ty);
1182             fiz0             = _mm_add_ps(fiz0,tz);
1183
1184             fjx0             = _mm_add_ps(fjx0,tx);
1185             fjy0             = _mm_add_ps(fjy0,ty);
1186             fjz0             = _mm_add_ps(fjz0,tz);
1187
1188             /**************************
1189              * CALCULATE INTERACTIONS *
1190              **************************/
1191
1192             r10              = _mm_mul_ps(rsq10,rinv10);
1193             r10              = _mm_andnot_ps(dummy_mask,r10);
1194
1195             /* Compute parameters for interactions between i and j atoms */
1196             qq10             = _mm_mul_ps(iq1,jq0);
1197
1198             /* Calculate table index by multiplying r with table scale and truncate to integer */
1199             rt               = _mm_mul_ps(r10,vftabscale);
1200             vfitab           = _mm_cvttps_epi32(rt);
1201             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1202             vfitab           = _mm_slli_epi32(vfitab,2);
1203
1204             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1205             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1206             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1207             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1208             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1209             _MM_TRANSPOSE4_PS(Y,F,G,H);
1210             Heps             = _mm_mul_ps(vfeps,H);
1211             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1212             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1213             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1214
1215             fscal            = felec;
1216
1217             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1218
1219             /* Calculate temporary vectorial force */
1220             tx               = _mm_mul_ps(fscal,dx10);
1221             ty               = _mm_mul_ps(fscal,dy10);
1222             tz               = _mm_mul_ps(fscal,dz10);
1223
1224             /* Update vectorial force */
1225             fix1             = _mm_add_ps(fix1,tx);
1226             fiy1             = _mm_add_ps(fiy1,ty);
1227             fiz1             = _mm_add_ps(fiz1,tz);
1228
1229             fjx0             = _mm_add_ps(fjx0,tx);
1230             fjy0             = _mm_add_ps(fjy0,ty);
1231             fjz0             = _mm_add_ps(fjz0,tz);
1232
1233             /**************************
1234              * CALCULATE INTERACTIONS *
1235              **************************/
1236
1237             r20              = _mm_mul_ps(rsq20,rinv20);
1238             r20              = _mm_andnot_ps(dummy_mask,r20);
1239
1240             /* Compute parameters for interactions between i and j atoms */
1241             qq20             = _mm_mul_ps(iq2,jq0);
1242
1243             /* Calculate table index by multiplying r with table scale and truncate to integer */
1244             rt               = _mm_mul_ps(r20,vftabscale);
1245             vfitab           = _mm_cvttps_epi32(rt);
1246             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1247             vfitab           = _mm_slli_epi32(vfitab,2);
1248
1249             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1250             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1251             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1252             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1253             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1254             _MM_TRANSPOSE4_PS(Y,F,G,H);
1255             Heps             = _mm_mul_ps(vfeps,H);
1256             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1257             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1258             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1259
1260             fscal            = felec;
1261
1262             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1263
1264             /* Calculate temporary vectorial force */
1265             tx               = _mm_mul_ps(fscal,dx20);
1266             ty               = _mm_mul_ps(fscal,dy20);
1267             tz               = _mm_mul_ps(fscal,dz20);
1268
1269             /* Update vectorial force */
1270             fix2             = _mm_add_ps(fix2,tx);
1271             fiy2             = _mm_add_ps(fiy2,ty);
1272             fiz2             = _mm_add_ps(fiz2,tz);
1273
1274             fjx0             = _mm_add_ps(fjx0,tx);
1275             fjy0             = _mm_add_ps(fjy0,ty);
1276             fjz0             = _mm_add_ps(fjz0,tz);
1277
1278             /**************************
1279              * CALCULATE INTERACTIONS *
1280              **************************/
1281
1282             r30              = _mm_mul_ps(rsq30,rinv30);
1283             r30              = _mm_andnot_ps(dummy_mask,r30);
1284
1285             /* Compute parameters for interactions between i and j atoms */
1286             qq30             = _mm_mul_ps(iq3,jq0);
1287
1288             /* Calculate table index by multiplying r with table scale and truncate to integer */
1289             rt               = _mm_mul_ps(r30,vftabscale);
1290             vfitab           = _mm_cvttps_epi32(rt);
1291             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1292             vfitab           = _mm_slli_epi32(vfitab,2);
1293
1294             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1295             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1296             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1297             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1298             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1299             _MM_TRANSPOSE4_PS(Y,F,G,H);
1300             Heps             = _mm_mul_ps(vfeps,H);
1301             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1302             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1303             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1304
1305             fscal            = felec;
1306
1307             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1308
1309             /* Calculate temporary vectorial force */
1310             tx               = _mm_mul_ps(fscal,dx30);
1311             ty               = _mm_mul_ps(fscal,dy30);
1312             tz               = _mm_mul_ps(fscal,dz30);
1313
1314             /* Update vectorial force */
1315             fix3             = _mm_add_ps(fix3,tx);
1316             fiy3             = _mm_add_ps(fiy3,ty);
1317             fiz3             = _mm_add_ps(fiz3,tz);
1318
1319             fjx0             = _mm_add_ps(fjx0,tx);
1320             fjy0             = _mm_add_ps(fjy0,ty);
1321             fjz0             = _mm_add_ps(fjz0,tz);
1322
1323             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1324             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1325             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1326             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1327
1328             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1329
1330             /* Inner loop uses 147 flops */
1331         }
1332
1333         /* End of innermost loop */
1334
1335         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1336                                               f+i_coord_offset,fshift+i_shift_offset);
1337
1338         /* Increment number of inner iterations */
1339         inneriter                  += j_index_end - j_index_start;
1340
1341         /* Outer loop uses 24 flops */
1342     }
1343
1344     /* Increment number of outer iterations */
1345     outeriter        += nri;
1346
1347     /* Update outer/inner flops */
1348
1349     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*147);
1350 }