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, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_single kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse4_1_single.h"
  50 #include "kernelutil_x86_sse4_1_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_VF_sse4_1_single
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_VF_sse4_1_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwioffset3;
  92     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  93     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  94     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  95     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  96     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  97     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  98     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  99     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     int              nvdwtype;
 102     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 106     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 107     __m128i          vfitab;
 108     __m128i          ifour       = _mm_set1_epi32(4);
 109     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 110     real             *vftab;
 111     __m128           dummy_mask,cutoff_mask;
 112     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 113     __m128           one     = _mm_set1_ps(1.0);
 114     __m128           two     = _mm_set1_ps(2.0);
 115     x                = xx[0];
 116     f                = ff[0];
 117
 118     nri              = nlist->nri;
 119     iinr             = nlist->iinr;
 120     jindex           = nlist->jindex;
 121     jjnr             = nlist->jjnr;
 122     shiftidx         = nlist->shift;
 123     gid              = nlist->gid;
 124     shiftvec         = fr->shift_vec[0];
 125     fshift           = fr->fshift[0];
 126     facel            = _mm_set1_ps(fr->epsfac);
 127     charge           = mdatoms->chargeA;
 128     nvdwtype         = fr->ntype;
 129     vdwparam         = fr->nbfp;
 130     vdwtype          = mdatoms->typeA;
 131
 132     vftab            = kernel_data->table_elec->data;
 133     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 134
 135     /* Setup water-specific parameters */
 136     inr              = nlist->iinr[0];
 137     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 138     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 139     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 140     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 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_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 173                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 174
 175         fix0             = _mm_setzero_ps();
 176         fiy0             = _mm_setzero_ps();
 177         fiz0             = _mm_setzero_ps();
 178         fix1             = _mm_setzero_ps();
 179         fiy1             = _mm_setzero_ps();
 180         fiz1             = _mm_setzero_ps();
 181         fix2             = _mm_setzero_ps();
 182         fiy2             = _mm_setzero_ps();
 183         fiz2             = _mm_setzero_ps();
 184         fix3             = _mm_setzero_ps();
 185         fiy3             = _mm_setzero_ps();
 186         fiz3             = _mm_setzero_ps();
 187
 188         /* Reset potential sums */
 189         velecsum         = _mm_setzero_ps();
 190         vvdwsum          = _mm_setzero_ps();
 191
 192         /* Start inner kernel loop */
 193         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 194         {
 195
 196             /* Get j neighbor index, and coordinate index */
 197             jnrA             = jjnr[jidx];
 198             jnrB             = jjnr[jidx+1];
 199             jnrC             = jjnr[jidx+2];
 200             jnrD             = jjnr[jidx+3];
 201             j_coord_offsetA  = DIM*jnrA;
 202             j_coord_offsetB  = DIM*jnrB;
 203             j_coord_offsetC  = DIM*jnrC;
 204             j_coord_offsetD  = DIM*jnrD;
 205
 206             /* load j atom coordinates */
 207             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 208                                               x+j_coord_offsetC,x+j_coord_offsetD,
 209                                               &jx0,&jy0,&jz0);
 210
 211             /* Calculate displacement vector */
 212             dx00             = _mm_sub_ps(ix0,jx0);
 213             dy00             = _mm_sub_ps(iy0,jy0);
 214             dz00             = _mm_sub_ps(iz0,jz0);
 215             dx10             = _mm_sub_ps(ix1,jx0);
 216             dy10             = _mm_sub_ps(iy1,jy0);
 217             dz10             = _mm_sub_ps(iz1,jz0);
 218             dx20             = _mm_sub_ps(ix2,jx0);
 219             dy20             = _mm_sub_ps(iy2,jy0);
 220             dz20             = _mm_sub_ps(iz2,jz0);
 221             dx30             = _mm_sub_ps(ix3,jx0);
 222             dy30             = _mm_sub_ps(iy3,jy0);
 223             dz30             = _mm_sub_ps(iz3,jz0);
 224
 225             /* Calculate squared distance and things based on it */
 226             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 227             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 228             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 229             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 230
 231             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 232             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 233             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 234
 235             rinvsq00         = gmx_mm_inv_ps(rsq00);
 236
 237             /* Load parameters for j particles */
 238             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 239                                                               charge+jnrC+0,charge+jnrD+0);
 240             vdwjidx0A        = 2*vdwtype[jnrA+0];
 241             vdwjidx0B        = 2*vdwtype[jnrB+0];
 242             vdwjidx0C        = 2*vdwtype[jnrC+0];
 243             vdwjidx0D        = 2*vdwtype[jnrD+0];
 244
 245             fjx0             = _mm_setzero_ps();
 246             fjy0             = _mm_setzero_ps();
 247             fjz0             = _mm_setzero_ps();
 248
 249             /**************************
 250              * CALCULATE INTERACTIONS *
 251              **************************/
 252
 253             /* Compute parameters for interactions between i and j atoms */
 254             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 255                                          vdwparam+vdwioffset0+vdwjidx0B,
 256                                          vdwparam+vdwioffset0+vdwjidx0C,
 257                                          vdwparam+vdwioffset0+vdwjidx0D,
 258                                          &c6_00,&c12_00);
 259
 260             /* LENNARD-JONES DISPERSION/REPULSION */
 261
 262             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 263             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 264             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 265             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 266             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 267
 268             /* Update potential sum for this i atom from the interaction with this j atom. */
 269             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 270
 271             fscal            = fvdw;
 272
 273             /* Calculate temporary vectorial force */
 274             tx               = _mm_mul_ps(fscal,dx00);
 275             ty               = _mm_mul_ps(fscal,dy00);
 276             tz               = _mm_mul_ps(fscal,dz00);
 277
 278             /* Update vectorial force */
 279             fix0             = _mm_add_ps(fix0,tx);
 280             fiy0             = _mm_add_ps(fiy0,ty);
 281             fiz0             = _mm_add_ps(fiz0,tz);
 282
 283             fjx0             = _mm_add_ps(fjx0,tx);
 284             fjy0             = _mm_add_ps(fjy0,ty);
 285             fjz0             = _mm_add_ps(fjz0,tz);
 286
 287             /**************************
 288              * CALCULATE INTERACTIONS *
 289              **************************/
 290
 291             r10              = _mm_mul_ps(rsq10,rinv10);
 292
 293             /* Compute parameters for interactions between i and j atoms */
 294             qq10             = _mm_mul_ps(iq1,jq0);
 295
 296             /* Calculate table index by multiplying r with table scale and truncate to integer */
 297             rt               = _mm_mul_ps(r10,vftabscale);
 298             vfitab           = _mm_cvttps_epi32(rt);
 299             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 300             vfitab           = _mm_slli_epi32(vfitab,2);
 301
 302             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 303             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 304             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 305             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 306             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 307             _MM_TRANSPOSE4_PS(Y,F,G,H);
 308             Heps             = _mm_mul_ps(vfeps,H);
 309             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 310             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 311             velec            = _mm_mul_ps(qq10,VV);
 312             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 313             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 314
 315             /* Update potential sum for this i atom from the interaction with this j atom. */
 316             velecsum         = _mm_add_ps(velecsum,velec);
 317
 318             fscal            = felec;
 319
 320             /* Calculate temporary vectorial force */
 321             tx               = _mm_mul_ps(fscal,dx10);
 322             ty               = _mm_mul_ps(fscal,dy10);
 323             tz               = _mm_mul_ps(fscal,dz10);
 324
 325             /* Update vectorial force */
 326             fix1             = _mm_add_ps(fix1,tx);
 327             fiy1             = _mm_add_ps(fiy1,ty);
 328             fiz1             = _mm_add_ps(fiz1,tz);
 329
 330             fjx0             = _mm_add_ps(fjx0,tx);
 331             fjy0             = _mm_add_ps(fjy0,ty);
 332             fjz0             = _mm_add_ps(fjz0,tz);
 333
 334             /**************************
 335              * CALCULATE INTERACTIONS *
 336              **************************/
 337
 338             r20              = _mm_mul_ps(rsq20,rinv20);
 339
 340             /* Compute parameters for interactions between i and j atoms */
 341             qq20             = _mm_mul_ps(iq2,jq0);
 342
 343             /* Calculate table index by multiplying r with table scale and truncate to integer */
 344             rt               = _mm_mul_ps(r20,vftabscale);
 345             vfitab           = _mm_cvttps_epi32(rt);
 346             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 347             vfitab           = _mm_slli_epi32(vfitab,2);
 348
 349             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 350             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 351             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 352             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 353             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 354             _MM_TRANSPOSE4_PS(Y,F,G,H);
 355             Heps             = _mm_mul_ps(vfeps,H);
 356             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 357             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 358             velec            = _mm_mul_ps(qq20,VV);
 359             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 360             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 361
 362             /* Update potential sum for this i atom from the interaction with this j atom. */
 363             velecsum         = _mm_add_ps(velecsum,velec);
 364
 365             fscal            = felec;
 366
 367             /* Calculate temporary vectorial force */
 368             tx               = _mm_mul_ps(fscal,dx20);
 369             ty               = _mm_mul_ps(fscal,dy20);
 370             tz               = _mm_mul_ps(fscal,dz20);
 371
 372             /* Update vectorial force */
 373             fix2             = _mm_add_ps(fix2,tx);
 374             fiy2             = _mm_add_ps(fiy2,ty);
 375             fiz2             = _mm_add_ps(fiz2,tz);
 376
 377             fjx0             = _mm_add_ps(fjx0,tx);
 378             fjy0             = _mm_add_ps(fjy0,ty);
 379             fjz0             = _mm_add_ps(fjz0,tz);
 380
 381             /**************************
 382              * CALCULATE INTERACTIONS *
 383              **************************/
 384
 385             r30              = _mm_mul_ps(rsq30,rinv30);
 386
 387             /* Compute parameters for interactions between i and j atoms */
 388             qq30             = _mm_mul_ps(iq3,jq0);
 389
 390             /* Calculate table index by multiplying r with table scale and truncate to integer */
 391             rt               = _mm_mul_ps(r30,vftabscale);
 392             vfitab           = _mm_cvttps_epi32(rt);
 393             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 394             vfitab           = _mm_slli_epi32(vfitab,2);
 395
 396             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 397             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 398             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 399             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 400             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 401             _MM_TRANSPOSE4_PS(Y,F,G,H);
 402             Heps             = _mm_mul_ps(vfeps,H);
 403             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 404             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 405             velec            = _mm_mul_ps(qq30,VV);
 406             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 407             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 408
 409             /* Update potential sum for this i atom from the interaction with this j atom. */
 410             velecsum         = _mm_add_ps(velecsum,velec);
 411
 412             fscal            = felec;
 413
 414             /* Calculate temporary vectorial force */
 415             tx               = _mm_mul_ps(fscal,dx30);
 416             ty               = _mm_mul_ps(fscal,dy30);
 417             tz               = _mm_mul_ps(fscal,dz30);
 418
 419             /* Update vectorial force */
 420             fix3             = _mm_add_ps(fix3,tx);
 421             fiy3             = _mm_add_ps(fiy3,ty);
 422             fiz3             = _mm_add_ps(fiz3,tz);
 423
 424             fjx0             = _mm_add_ps(fjx0,tx);
 425             fjy0             = _mm_add_ps(fjy0,ty);
 426             fjz0             = _mm_add_ps(fjz0,tz);
 427
 428             fjptrA             = f+j_coord_offsetA;
 429             fjptrB             = f+j_coord_offsetB;
 430             fjptrC             = f+j_coord_offsetC;
 431             fjptrD             = f+j_coord_offsetD;
 432
 433             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 434
 435             /* Inner loop uses 161 flops */
 436         }
 437
 438         if(jidx<j_index_end)
 439         {
 440
 441             /* Get j neighbor index, and coordinate index */
 442             jnrlistA         = jjnr[jidx];
 443             jnrlistB         = jjnr[jidx+1];
 444             jnrlistC         = jjnr[jidx+2];
 445             jnrlistD         = jjnr[jidx+3];
 446             /* Sign of each element will be negative for non-real atoms.
 447              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 448              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 449              */
 450             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 451             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 452             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 453             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 454             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 455             j_coord_offsetA  = DIM*jnrA;
 456             j_coord_offsetB  = DIM*jnrB;
 457             j_coord_offsetC  = DIM*jnrC;
 458             j_coord_offsetD  = DIM*jnrD;
 459
 460             /* load j atom coordinates */
 461             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 462                                               x+j_coord_offsetC,x+j_coord_offsetD,
 463                                               &jx0,&jy0,&jz0);
 464
 465             /* Calculate displacement vector */
 466             dx00             = _mm_sub_ps(ix0,jx0);
 467             dy00             = _mm_sub_ps(iy0,jy0);
 468             dz00             = _mm_sub_ps(iz0,jz0);
 469             dx10             = _mm_sub_ps(ix1,jx0);
 470             dy10             = _mm_sub_ps(iy1,jy0);
 471             dz10             = _mm_sub_ps(iz1,jz0);
 472             dx20             = _mm_sub_ps(ix2,jx0);
 473             dy20             = _mm_sub_ps(iy2,jy0);
 474             dz20             = _mm_sub_ps(iz2,jz0);
 475             dx30             = _mm_sub_ps(ix3,jx0);
 476             dy30             = _mm_sub_ps(iy3,jy0);
 477             dz30             = _mm_sub_ps(iz3,jz0);
 478
 479             /* Calculate squared distance and things based on it */
 480             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 481             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 482             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 483             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 484
 485             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 486             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 487             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 488
 489             rinvsq00         = gmx_mm_inv_ps(rsq00);
 490
 491             /* Load parameters for j particles */
 492             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 493                                                               charge+jnrC+0,charge+jnrD+0);
 494             vdwjidx0A        = 2*vdwtype[jnrA+0];
 495             vdwjidx0B        = 2*vdwtype[jnrB+0];
 496             vdwjidx0C        = 2*vdwtype[jnrC+0];
 497             vdwjidx0D        = 2*vdwtype[jnrD+0];
 498
 499             fjx0             = _mm_setzero_ps();
 500             fjy0             = _mm_setzero_ps();
 501             fjz0             = _mm_setzero_ps();
 502
 503             /**************************
 504              * CALCULATE INTERACTIONS *
 505              **************************/
 506
 507             /* Compute parameters for interactions between i and j atoms */
 508             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 509                                          vdwparam+vdwioffset0+vdwjidx0B,
 510                                          vdwparam+vdwioffset0+vdwjidx0C,
 511                                          vdwparam+vdwioffset0+vdwjidx0D,
 512                                          &c6_00,&c12_00);
 513
 514             /* LENNARD-JONES DISPERSION/REPULSION */
 515
 516             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 517             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 518             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 519             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 520             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 521
 522             /* Update potential sum for this i atom from the interaction with this j atom. */
 523             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 524             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 525
 526             fscal            = fvdw;
 527
 528             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 529
 530             /* Calculate temporary vectorial force */
 531             tx               = _mm_mul_ps(fscal,dx00);
 532             ty               = _mm_mul_ps(fscal,dy00);
 533             tz               = _mm_mul_ps(fscal,dz00);
 534
 535             /* Update vectorial force */
 536             fix0             = _mm_add_ps(fix0,tx);
 537             fiy0             = _mm_add_ps(fiy0,ty);
 538             fiz0             = _mm_add_ps(fiz0,tz);
 539
 540             fjx0             = _mm_add_ps(fjx0,tx);
 541             fjy0             = _mm_add_ps(fjy0,ty);
 542             fjz0             = _mm_add_ps(fjz0,tz);
 543
 544             /**************************
 545              * CALCULATE INTERACTIONS *
 546              **************************/
 547
 548             r10              = _mm_mul_ps(rsq10,rinv10);
 549             r10              = _mm_andnot_ps(dummy_mask,r10);
 550
 551             /* Compute parameters for interactions between i and j atoms */
 552             qq10             = _mm_mul_ps(iq1,jq0);
 553
 554             /* Calculate table index by multiplying r with table scale and truncate to integer */
 555             rt               = _mm_mul_ps(r10,vftabscale);
 556             vfitab           = _mm_cvttps_epi32(rt);
 557             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 558             vfitab           = _mm_slli_epi32(vfitab,2);
 559
 560             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 561             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 562             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 563             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 564             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 565             _MM_TRANSPOSE4_PS(Y,F,G,H);
 566             Heps             = _mm_mul_ps(vfeps,H);
 567             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 568             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 569             velec            = _mm_mul_ps(qq10,VV);
 570             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 571             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 572
 573             /* Update potential sum for this i atom from the interaction with this j atom. */
 574             velec            = _mm_andnot_ps(dummy_mask,velec);
 575             velecsum         = _mm_add_ps(velecsum,velec);
 576
 577             fscal            = felec;
 578
 579             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 580
 581             /* Calculate temporary vectorial force */
 582             tx               = _mm_mul_ps(fscal,dx10);
 583             ty               = _mm_mul_ps(fscal,dy10);
 584             tz               = _mm_mul_ps(fscal,dz10);
 585
 586             /* Update vectorial force */
 587             fix1             = _mm_add_ps(fix1,tx);
 588             fiy1             = _mm_add_ps(fiy1,ty);
 589             fiz1             = _mm_add_ps(fiz1,tz);
 590
 591             fjx0             = _mm_add_ps(fjx0,tx);
 592             fjy0             = _mm_add_ps(fjy0,ty);
 593             fjz0             = _mm_add_ps(fjz0,tz);
 594
 595             /**************************
 596              * CALCULATE INTERACTIONS *
 597              **************************/
 598
 599             r20              = _mm_mul_ps(rsq20,rinv20);
 600             r20              = _mm_andnot_ps(dummy_mask,r20);
 601
 602             /* Compute parameters for interactions between i and j atoms */
 603             qq20             = _mm_mul_ps(iq2,jq0);
 604
 605             /* Calculate table index by multiplying r with table scale and truncate to integer */
 606             rt               = _mm_mul_ps(r20,vftabscale);
 607             vfitab           = _mm_cvttps_epi32(rt);
 608             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 609             vfitab           = _mm_slli_epi32(vfitab,2);
 610
 611             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 612             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 613             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 614             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 615             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 616             _MM_TRANSPOSE4_PS(Y,F,G,H);
 617             Heps             = _mm_mul_ps(vfeps,H);
 618             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 619             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 620             velec            = _mm_mul_ps(qq20,VV);
 621             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 622             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 623
 624             /* Update potential sum for this i atom from the interaction with this j atom. */
 625             velec            = _mm_andnot_ps(dummy_mask,velec);
 626             velecsum         = _mm_add_ps(velecsum,velec);
 627
 628             fscal            = felec;
 629
 630             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 631
 632             /* Calculate temporary vectorial force */
 633             tx               = _mm_mul_ps(fscal,dx20);
 634             ty               = _mm_mul_ps(fscal,dy20);
 635             tz               = _mm_mul_ps(fscal,dz20);
 636
 637             /* Update vectorial force */
 638             fix2             = _mm_add_ps(fix2,tx);
 639             fiy2             = _mm_add_ps(fiy2,ty);
 640             fiz2             = _mm_add_ps(fiz2,tz);
 641
 642             fjx0             = _mm_add_ps(fjx0,tx);
 643             fjy0             = _mm_add_ps(fjy0,ty);
 644             fjz0             = _mm_add_ps(fjz0,tz);
 645
 646             /**************************
 647              * CALCULATE INTERACTIONS *
 648              **************************/
 649
 650             r30              = _mm_mul_ps(rsq30,rinv30);
 651             r30              = _mm_andnot_ps(dummy_mask,r30);
 652
 653             /* Compute parameters for interactions between i and j atoms */
 654             qq30             = _mm_mul_ps(iq3,jq0);
 655
 656             /* Calculate table index by multiplying r with table scale and truncate to integer */
 657             rt               = _mm_mul_ps(r30,vftabscale);
 658             vfitab           = _mm_cvttps_epi32(rt);
 659             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 660             vfitab           = _mm_slli_epi32(vfitab,2);
 661
 662             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 663             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 664             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 665             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 666             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 667             _MM_TRANSPOSE4_PS(Y,F,G,H);
 668             Heps             = _mm_mul_ps(vfeps,H);
 669             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 670             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 671             velec            = _mm_mul_ps(qq30,VV);
 672             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 673             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 674
 675             /* Update potential sum for this i atom from the interaction with this j atom. */
 676             velec            = _mm_andnot_ps(dummy_mask,velec);
 677             velecsum         = _mm_add_ps(velecsum,velec);
 678
 679             fscal            = felec;
 680
 681             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 682
 683             /* Calculate temporary vectorial force */
 684             tx               = _mm_mul_ps(fscal,dx30);
 685             ty               = _mm_mul_ps(fscal,dy30);
 686             tz               = _mm_mul_ps(fscal,dz30);
 687
 688             /* Update vectorial force */
 689             fix3             = _mm_add_ps(fix3,tx);
 690             fiy3             = _mm_add_ps(fiy3,ty);
 691             fiz3             = _mm_add_ps(fiz3,tz);
 692
 693             fjx0             = _mm_add_ps(fjx0,tx);
 694             fjy0             = _mm_add_ps(fjy0,ty);
 695             fjz0             = _mm_add_ps(fjz0,tz);
 696
 697             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 698             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 699             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 700             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 701
 702             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 703
 704             /* Inner loop uses 164 flops */
 705         }
 706
 707         /* End of innermost loop */
 708
 709         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 710                                               f+i_coord_offset,fshift+i_shift_offset);
 711
 712         ggid                        = gid[iidx];
 713         /* Update potential energies */
 714         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 715         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 716
 717         /* Increment number of inner iterations */
 718         inneriter                  += j_index_end - j_index_start;
 719
 720         /* Outer loop uses 26 flops */
 721     }
 722
 723     /* Increment number of outer iterations */
 724     outeriter        += nri;
 725
 726     /* Update outer/inner flops */
 727
 728     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*164);
 729 }
 730 /*
 731  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_sse4_1_single
 732  * Electrostatics interaction: CubicSplineTable
 733  * VdW interaction:            LennardJones
 734  * Geometry:                   Water4-Particle
 735  * Calculate force/pot:        Force
 736  */
 737 void
 738 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_sse4_1_single
 739                     (t_nblist                    * gmx_restrict       nlist,
 740                      rvec                        * gmx_restrict          xx,
 741                      rvec                        * gmx_restrict          ff,
 742                      t_forcerec                  * gmx_restrict          fr,
 743                      t_mdatoms                   * gmx_restrict     mdatoms,
 744                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 745                      t_nrnb                      * gmx_restrict        nrnb)
 746 {
 747     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 748      * just 0 for non-waters.
 749      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 750      * jnr indices corresponding to data put in the four positions in the SIMD register.
 751      */
 752     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 753     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 754     int              jnrA,jnrB,jnrC,jnrD;
 755     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 756     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 757     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 758     real             rcutoff_scalar;
 759     real             *shiftvec,*fshift,*x,*f;
 760     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 761     real             scratch[4*DIM];
 762     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 763     int              vdwioffset0;
 764     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 765     int              vdwioffset1;
 766     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 767     int              vdwioffset2;
 768     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 769     int              vdwioffset3;
 770     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 771     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 772     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 773     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 774     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 775     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 776     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 777     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 778     real             *charge;
 779     int              nvdwtype;
 780     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 781     int              *vdwtype;
 782     real             *vdwparam;
 783     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 784     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 785     __m128i          vfitab;
 786     __m128i          ifour       = _mm_set1_epi32(4);
 787     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 788     real             *vftab;
 789     __m128           dummy_mask,cutoff_mask;
 790     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 791     __m128           one     = _mm_set1_ps(1.0);
 792     __m128           two     = _mm_set1_ps(2.0);
 793     x                = xx[0];
 794     f                = ff[0];
 795
 796     nri              = nlist->nri;
 797     iinr             = nlist->iinr;
 798     jindex           = nlist->jindex;
 799     jjnr             = nlist->jjnr;
 800     shiftidx         = nlist->shift;
 801     gid              = nlist->gid;
 802     shiftvec         = fr->shift_vec[0];
 803     fshift           = fr->fshift[0];
 804     facel            = _mm_set1_ps(fr->epsfac);
 805     charge           = mdatoms->chargeA;
 806     nvdwtype         = fr->ntype;
 807     vdwparam         = fr->nbfp;
 808     vdwtype          = mdatoms->typeA;
 809
 810     vftab            = kernel_data->table_elec->data;
 811     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 812
 813     /* Setup water-specific parameters */
 814     inr              = nlist->iinr[0];
 815     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 816     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 817     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 818     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 819
 820     /* Avoid stupid compiler warnings */
 821     jnrA = jnrB = jnrC = jnrD = 0;
 822     j_coord_offsetA = 0;
 823     j_coord_offsetB = 0;
 824     j_coord_offsetC = 0;
 825     j_coord_offsetD = 0;
 826
 827     outeriter        = 0;
 828     inneriter        = 0;
 829
 830     for(iidx=0;iidx<4*DIM;iidx++)
 831     {
 832         scratch[iidx] = 0.0;
 833     }
 834
 835     /* Start outer loop over neighborlists */
 836     for(iidx=0; iidx<nri; iidx++)
 837     {
 838         /* Load shift vector for this list */
 839         i_shift_offset   = DIM*shiftidx[iidx];
 840
 841         /* Load limits for loop over neighbors */
 842         j_index_start    = jindex[iidx];
 843         j_index_end      = jindex[iidx+1];
 844
 845         /* Get outer coordinate index */
 846         inr              = iinr[iidx];
 847         i_coord_offset   = DIM*inr;
 848
 849         /* Load i particle coords and add shift vector */
 850         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 851                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 852
 853         fix0             = _mm_setzero_ps();
 854         fiy0             = _mm_setzero_ps();
 855         fiz0             = _mm_setzero_ps();
 856         fix1             = _mm_setzero_ps();
 857         fiy1             = _mm_setzero_ps();
 858         fiz1             = _mm_setzero_ps();
 859         fix2             = _mm_setzero_ps();
 860         fiy2             = _mm_setzero_ps();
 861         fiz2             = _mm_setzero_ps();
 862         fix3             = _mm_setzero_ps();
 863         fiy3             = _mm_setzero_ps();
 864         fiz3             = _mm_setzero_ps();
 865
 866         /* Start inner kernel loop */
 867         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 868         {
 869
 870             /* Get j neighbor index, and coordinate index */
 871             jnrA             = jjnr[jidx];
 872             jnrB             = jjnr[jidx+1];
 873             jnrC             = jjnr[jidx+2];
 874             jnrD             = jjnr[jidx+3];
 875             j_coord_offsetA  = DIM*jnrA;
 876             j_coord_offsetB  = DIM*jnrB;
 877             j_coord_offsetC  = DIM*jnrC;
 878             j_coord_offsetD  = DIM*jnrD;
 879
 880             /* load j atom coordinates */
 881             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 882                                               x+j_coord_offsetC,x+j_coord_offsetD,
 883                                               &jx0,&jy0,&jz0);
 884
 885             /* Calculate displacement vector */
 886             dx00             = _mm_sub_ps(ix0,jx0);
 887             dy00             = _mm_sub_ps(iy0,jy0);
 888             dz00             = _mm_sub_ps(iz0,jz0);
 889             dx10             = _mm_sub_ps(ix1,jx0);
 890             dy10             = _mm_sub_ps(iy1,jy0);
 891             dz10             = _mm_sub_ps(iz1,jz0);
 892             dx20             = _mm_sub_ps(ix2,jx0);
 893             dy20             = _mm_sub_ps(iy2,jy0);
 894             dz20             = _mm_sub_ps(iz2,jz0);
 895             dx30             = _mm_sub_ps(ix3,jx0);
 896             dy30             = _mm_sub_ps(iy3,jy0);
 897             dz30             = _mm_sub_ps(iz3,jz0);
 898
 899             /* Calculate squared distance and things based on it */
 900             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 901             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 902             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 903             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 904
 905             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 906             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 907             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 908
 909             rinvsq00         = gmx_mm_inv_ps(rsq00);
 910
 911             /* Load parameters for j particles */
 912             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 913                                                               charge+jnrC+0,charge+jnrD+0);
 914             vdwjidx0A        = 2*vdwtype[jnrA+0];
 915             vdwjidx0B        = 2*vdwtype[jnrB+0];
 916             vdwjidx0C        = 2*vdwtype[jnrC+0];
 917             vdwjidx0D        = 2*vdwtype[jnrD+0];
 918
 919             fjx0             = _mm_setzero_ps();
 920             fjy0             = _mm_setzero_ps();
 921             fjz0             = _mm_setzero_ps();
 922
 923             /**************************
 924              * CALCULATE INTERACTIONS *
 925              **************************/
 926
 927             /* Compute parameters for interactions between i and j atoms */
 928             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 929                                          vdwparam+vdwioffset0+vdwjidx0B,
 930                                          vdwparam+vdwioffset0+vdwjidx0C,
 931                                          vdwparam+vdwioffset0+vdwjidx0D,
 932                                          &c6_00,&c12_00);
 933
 934             /* LENNARD-JONES DISPERSION/REPULSION */
 935
 936             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 937             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 938
 939             fscal            = fvdw;
 940
 941             /* Calculate temporary vectorial force */
 942             tx               = _mm_mul_ps(fscal,dx00);
 943             ty               = _mm_mul_ps(fscal,dy00);
 944             tz               = _mm_mul_ps(fscal,dz00);
 945
 946             /* Update vectorial force */
 947             fix0             = _mm_add_ps(fix0,tx);
 948             fiy0             = _mm_add_ps(fiy0,ty);
 949             fiz0             = _mm_add_ps(fiz0,tz);
 950
 951             fjx0             = _mm_add_ps(fjx0,tx);
 952             fjy0             = _mm_add_ps(fjy0,ty);
 953             fjz0             = _mm_add_ps(fjz0,tz);
 954
 955             /**************************
 956              * CALCULATE INTERACTIONS *
 957              **************************/
 958
 959             r10              = _mm_mul_ps(rsq10,rinv10);
 960
 961             /* Compute parameters for interactions between i and j atoms */
 962             qq10             = _mm_mul_ps(iq1,jq0);
 963
 964             /* Calculate table index by multiplying r with table scale and truncate to integer */
 965             rt               = _mm_mul_ps(r10,vftabscale);
 966             vfitab           = _mm_cvttps_epi32(rt);
 967             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 968             vfitab           = _mm_slli_epi32(vfitab,2);
 969
 970             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 971             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 972             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 973             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 974             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 975             _MM_TRANSPOSE4_PS(Y,F,G,H);
 976             Heps             = _mm_mul_ps(vfeps,H);
 977             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 978             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 979             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 980
 981             fscal            = felec;
 982
 983             /* Calculate temporary vectorial force */
 984             tx               = _mm_mul_ps(fscal,dx10);
 985             ty               = _mm_mul_ps(fscal,dy10);
 986             tz               = _mm_mul_ps(fscal,dz10);
 987
 988             /* Update vectorial force */
 989             fix1             = _mm_add_ps(fix1,tx);
 990             fiy1             = _mm_add_ps(fiy1,ty);
 991             fiz1             = _mm_add_ps(fiz1,tz);
 992
 993             fjx0             = _mm_add_ps(fjx0,tx);
 994             fjy0             = _mm_add_ps(fjy0,ty);
 995             fjz0             = _mm_add_ps(fjz0,tz);
 996
 997             /**************************
 998              * CALCULATE INTERACTIONS *
 999              **************************/
1000
1001             r20              = _mm_mul_ps(rsq20,rinv20);
1002
1003             /* Compute parameters for interactions between i and j atoms */
1004             qq20             = _mm_mul_ps(iq2,jq0);
1005
1006             /* Calculate table index by multiplying r with table scale and truncate to integer */
1007             rt               = _mm_mul_ps(r20,vftabscale);
1008             vfitab           = _mm_cvttps_epi32(rt);
1009             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1010             vfitab           = _mm_slli_epi32(vfitab,2);
1011
1012             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1013             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1014             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1015             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1016             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1017             _MM_TRANSPOSE4_PS(Y,F,G,H);
1018             Heps             = _mm_mul_ps(vfeps,H);
1019             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1020             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1021             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1022
1023             fscal            = felec;
1024
1025             /* Calculate temporary vectorial force */
1026             tx               = _mm_mul_ps(fscal,dx20);
1027             ty               = _mm_mul_ps(fscal,dy20);
1028             tz               = _mm_mul_ps(fscal,dz20);
1029
1030             /* Update vectorial force */
1031             fix2             = _mm_add_ps(fix2,tx);
1032             fiy2             = _mm_add_ps(fiy2,ty);
1033             fiz2             = _mm_add_ps(fiz2,tz);
1034
1035             fjx0             = _mm_add_ps(fjx0,tx);
1036             fjy0             = _mm_add_ps(fjy0,ty);
1037             fjz0             = _mm_add_ps(fjz0,tz);
1038
1039             /**************************
1040              * CALCULATE INTERACTIONS *
1041              **************************/
1042
1043             r30              = _mm_mul_ps(rsq30,rinv30);
1044
1045             /* Compute parameters for interactions between i and j atoms */
1046             qq30             = _mm_mul_ps(iq3,jq0);
1047
1048             /* Calculate table index by multiplying r with table scale and truncate to integer */
1049             rt               = _mm_mul_ps(r30,vftabscale);
1050             vfitab           = _mm_cvttps_epi32(rt);
1051             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1052             vfitab           = _mm_slli_epi32(vfitab,2);
1053
1054             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1055             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1056             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1057             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1058             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1059             _MM_TRANSPOSE4_PS(Y,F,G,H);
1060             Heps             = _mm_mul_ps(vfeps,H);
1061             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1062             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1063             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1064
1065             fscal            = felec;
1066
1067             /* Calculate temporary vectorial force */
1068             tx               = _mm_mul_ps(fscal,dx30);
1069             ty               = _mm_mul_ps(fscal,dy30);
1070             tz               = _mm_mul_ps(fscal,dz30);
1071
1072             /* Update vectorial force */
1073             fix3             = _mm_add_ps(fix3,tx);
1074             fiy3             = _mm_add_ps(fiy3,ty);
1075             fiz3             = _mm_add_ps(fiz3,tz);
1076
1077             fjx0             = _mm_add_ps(fjx0,tx);
1078             fjy0             = _mm_add_ps(fjy0,ty);
1079             fjz0             = _mm_add_ps(fjz0,tz);
1080
1081             fjptrA             = f+j_coord_offsetA;
1082             fjptrB             = f+j_coord_offsetB;
1083             fjptrC             = f+j_coord_offsetC;
1084             fjptrD             = f+j_coord_offsetD;
1085
1086             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1087
1088             /* Inner loop uses 144 flops */
1089         }
1090
1091         if(jidx<j_index_end)
1092         {
1093
1094             /* Get j neighbor index, and coordinate index */
1095             jnrlistA         = jjnr[jidx];
1096             jnrlistB         = jjnr[jidx+1];
1097             jnrlistC         = jjnr[jidx+2];
1098             jnrlistD         = jjnr[jidx+3];
1099             /* Sign of each element will be negative for non-real atoms.
1100              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1101              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1102              */
1103             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1104             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1105             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1106             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1107             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1108             j_coord_offsetA  = DIM*jnrA;
1109             j_coord_offsetB  = DIM*jnrB;
1110             j_coord_offsetC  = DIM*jnrC;
1111             j_coord_offsetD  = DIM*jnrD;
1112
1113             /* load j atom coordinates */
1114             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1115                                               x+j_coord_offsetC,x+j_coord_offsetD,
1116                                               &jx0,&jy0,&jz0);
1117
1118             /* Calculate displacement vector */
1119             dx00             = _mm_sub_ps(ix0,jx0);
1120             dy00             = _mm_sub_ps(iy0,jy0);
1121             dz00             = _mm_sub_ps(iz0,jz0);
1122             dx10             = _mm_sub_ps(ix1,jx0);
1123             dy10             = _mm_sub_ps(iy1,jy0);
1124             dz10             = _mm_sub_ps(iz1,jz0);
1125             dx20             = _mm_sub_ps(ix2,jx0);
1126             dy20             = _mm_sub_ps(iy2,jy0);
1127             dz20             = _mm_sub_ps(iz2,jz0);
1128             dx30             = _mm_sub_ps(ix3,jx0);
1129             dy30             = _mm_sub_ps(iy3,jy0);
1130             dz30             = _mm_sub_ps(iz3,jz0);
1131
1132             /* Calculate squared distance and things based on it */
1133             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1134             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1135             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1136             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1137
1138             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1139             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1140             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1141
1142             rinvsq00         = gmx_mm_inv_ps(rsq00);
1143
1144             /* Load parameters for j particles */
1145             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1146                                                               charge+jnrC+0,charge+jnrD+0);
1147             vdwjidx0A        = 2*vdwtype[jnrA+0];
1148             vdwjidx0B        = 2*vdwtype[jnrB+0];
1149             vdwjidx0C        = 2*vdwtype[jnrC+0];
1150             vdwjidx0D        = 2*vdwtype[jnrD+0];
1151
1152             fjx0             = _mm_setzero_ps();
1153             fjy0             = _mm_setzero_ps();
1154             fjz0             = _mm_setzero_ps();
1155
1156             /**************************
1157              * CALCULATE INTERACTIONS *
1158              **************************/
1159
1160             /* Compute parameters for interactions between i and j atoms */
1161             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1162                                          vdwparam+vdwioffset0+vdwjidx0B,
1163                                          vdwparam+vdwioffset0+vdwjidx0C,
1164                                          vdwparam+vdwioffset0+vdwjidx0D,
1165                                          &c6_00,&c12_00);
1166
1167             /* LENNARD-JONES DISPERSION/REPULSION */
1168
1169             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1170             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1171
1172             fscal            = fvdw;
1173
1174             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1175
1176             /* Calculate temporary vectorial force */
1177             tx               = _mm_mul_ps(fscal,dx00);
1178             ty               = _mm_mul_ps(fscal,dy00);
1179             tz               = _mm_mul_ps(fscal,dz00);
1180
1181             /* Update vectorial force */
1182             fix0             = _mm_add_ps(fix0,tx);
1183             fiy0             = _mm_add_ps(fiy0,ty);
1184             fiz0             = _mm_add_ps(fiz0,tz);
1185
1186             fjx0             = _mm_add_ps(fjx0,tx);
1187             fjy0             = _mm_add_ps(fjy0,ty);
1188             fjz0             = _mm_add_ps(fjz0,tz);
1189
1190             /**************************
1191              * CALCULATE INTERACTIONS *
1192              **************************/
1193
1194             r10              = _mm_mul_ps(rsq10,rinv10);
1195             r10              = _mm_andnot_ps(dummy_mask,r10);
1196
1197             /* Compute parameters for interactions between i and j atoms */
1198             qq10             = _mm_mul_ps(iq1,jq0);
1199
1200             /* Calculate table index by multiplying r with table scale and truncate to integer */
1201             rt               = _mm_mul_ps(r10,vftabscale);
1202             vfitab           = _mm_cvttps_epi32(rt);
1203             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1204             vfitab           = _mm_slli_epi32(vfitab,2);
1205
1206             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1207             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1208             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1209             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1210             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1211             _MM_TRANSPOSE4_PS(Y,F,G,H);
1212             Heps             = _mm_mul_ps(vfeps,H);
1213             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1214             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1215             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1216
1217             fscal            = felec;
1218
1219             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1220
1221             /* Calculate temporary vectorial force */
1222             tx               = _mm_mul_ps(fscal,dx10);
1223             ty               = _mm_mul_ps(fscal,dy10);
1224             tz               = _mm_mul_ps(fscal,dz10);
1225
1226             /* Update vectorial force */
1227             fix1             = _mm_add_ps(fix1,tx);
1228             fiy1             = _mm_add_ps(fiy1,ty);
1229             fiz1             = _mm_add_ps(fiz1,tz);
1230
1231             fjx0             = _mm_add_ps(fjx0,tx);
1232             fjy0             = _mm_add_ps(fjy0,ty);
1233             fjz0             = _mm_add_ps(fjz0,tz);
1234
1235             /**************************
1236              * CALCULATE INTERACTIONS *
1237              **************************/
1238
1239             r20              = _mm_mul_ps(rsq20,rinv20);
1240             r20              = _mm_andnot_ps(dummy_mask,r20);
1241
1242             /* Compute parameters for interactions between i and j atoms */
1243             qq20             = _mm_mul_ps(iq2,jq0);
1244
1245             /* Calculate table index by multiplying r with table scale and truncate to integer */
1246             rt               = _mm_mul_ps(r20,vftabscale);
1247             vfitab           = _mm_cvttps_epi32(rt);
1248             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1249             vfitab           = _mm_slli_epi32(vfitab,2);
1250
1251             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1252             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1253             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1254             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1255             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1256             _MM_TRANSPOSE4_PS(Y,F,G,H);
1257             Heps             = _mm_mul_ps(vfeps,H);
1258             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1259             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1260             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1261
1262             fscal            = felec;
1263
1264             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1265
1266             /* Calculate temporary vectorial force */
1267             tx               = _mm_mul_ps(fscal,dx20);
1268             ty               = _mm_mul_ps(fscal,dy20);
1269             tz               = _mm_mul_ps(fscal,dz20);
1270
1271             /* Update vectorial force */
1272             fix2             = _mm_add_ps(fix2,tx);
1273             fiy2             = _mm_add_ps(fiy2,ty);
1274             fiz2             = _mm_add_ps(fiz2,tz);
1275
1276             fjx0             = _mm_add_ps(fjx0,tx);
1277             fjy0             = _mm_add_ps(fjy0,ty);
1278             fjz0             = _mm_add_ps(fjz0,tz);
1279
1280             /**************************
1281              * CALCULATE INTERACTIONS *
1282              **************************/
1283
1284             r30              = _mm_mul_ps(rsq30,rinv30);
1285             r30              = _mm_andnot_ps(dummy_mask,r30);
1286
1287             /* Compute parameters for interactions between i and j atoms */
1288             qq30             = _mm_mul_ps(iq3,jq0);
1289
1290             /* Calculate table index by multiplying r with table scale and truncate to integer */
1291             rt               = _mm_mul_ps(r30,vftabscale);
1292             vfitab           = _mm_cvttps_epi32(rt);
1293             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1294             vfitab           = _mm_slli_epi32(vfitab,2);
1295
1296             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1297             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1298             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1299             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1300             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1301             _MM_TRANSPOSE4_PS(Y,F,G,H);
1302             Heps             = _mm_mul_ps(vfeps,H);
1303             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1304             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1305             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1306
1307             fscal            = felec;
1308
1309             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1310
1311             /* Calculate temporary vectorial force */
1312             tx               = _mm_mul_ps(fscal,dx30);
1313             ty               = _mm_mul_ps(fscal,dy30);
1314             tz               = _mm_mul_ps(fscal,dz30);
1315
1316             /* Update vectorial force */
1317             fix3             = _mm_add_ps(fix3,tx);
1318             fiy3             = _mm_add_ps(fiy3,ty);
1319             fiz3             = _mm_add_ps(fiz3,tz);
1320
1321             fjx0             = _mm_add_ps(fjx0,tx);
1322             fjy0             = _mm_add_ps(fjy0,ty);
1323             fjz0             = _mm_add_ps(fjz0,tz);
1324
1325             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1326             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1327             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1328             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1329
1330             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1331
1332             /* Inner loop uses 147 flops */
1333         }
1334
1335         /* End of innermost loop */
1336
1337         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1338                                               f+i_coord_offset,fshift+i_shift_offset);
1339
1340         /* Increment number of inner iterations */
1341         inneriter                  += j_index_end - j_index_start;
1342
1343         /* Outer loop uses 24 flops */
1344     }
1345
1346     /* Increment number of outer iterations */
1347     outeriter        += nri;
1348
1349     /* Update outer/inner flops */
1350
1351     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*147);
1352 }