src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecCoul_VdwLJ_GeomW3P1_sse4_1_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_single.h"
  48 #include "kernelutil_x86_sse4_1_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW3P1_VF_sse4_1_single
  52  * Electrostatics interaction: Coulomb
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_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              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  92     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  93     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  94     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 101     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 102     __m128           dummy_mask,cutoff_mask;
 103     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 104     __m128           one     = _mm_set1_ps(1.0);
 105     __m128           two     = _mm_set1_ps(2.0);
 106     x                = xx[0];
 107     f                = ff[0];
 108
 109     nri              = nlist->nri;
 110     iinr             = nlist->iinr;
 111     jindex           = nlist->jindex;
 112     jjnr             = nlist->jjnr;
 113     shiftidx         = nlist->shift;
 114     gid              = nlist->gid;
 115     shiftvec         = fr->shift_vec[0];
 116     fshift           = fr->fshift[0];
 117     facel            = _mm_set1_ps(fr->epsfac);
 118     charge           = mdatoms->chargeA;
 119     nvdwtype         = fr->ntype;
 120     vdwparam         = fr->nbfp;
 121     vdwtype          = mdatoms->typeA;
 122
 123     /* Setup water-specific parameters */
 124     inr              = nlist->iinr[0];
 125     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 126     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 127     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 128     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 129
 130     /* Avoid stupid compiler warnings */
 131     jnrA = jnrB = jnrC = jnrD = 0;
 132     j_coord_offsetA = 0;
 133     j_coord_offsetB = 0;
 134     j_coord_offsetC = 0;
 135     j_coord_offsetD = 0;
 136
 137     outeriter        = 0;
 138     inneriter        = 0;
 139
 140     for(iidx=0;iidx<4*DIM;iidx++)
 141     {
 142         scratch[iidx] = 0.0;
 143     }
 144
 145     /* Start outer loop over neighborlists */
 146     for(iidx=0; iidx<nri; iidx++)
 147     {
 148         /* Load shift vector for this list */
 149         i_shift_offset   = DIM*shiftidx[iidx];
 150
 151         /* Load limits for loop over neighbors */
 152         j_index_start    = jindex[iidx];
 153         j_index_end      = jindex[iidx+1];
 154
 155         /* Get outer coordinate index */
 156         inr              = iinr[iidx];
 157         i_coord_offset   = DIM*inr;
 158
 159         /* Load i particle coords and add shift vector */
 160         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 161                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 162
 163         fix0             = _mm_setzero_ps();
 164         fiy0             = _mm_setzero_ps();
 165         fiz0             = _mm_setzero_ps();
 166         fix1             = _mm_setzero_ps();
 167         fiy1             = _mm_setzero_ps();
 168         fiz1             = _mm_setzero_ps();
 169         fix2             = _mm_setzero_ps();
 170         fiy2             = _mm_setzero_ps();
 171         fiz2             = _mm_setzero_ps();
 172
 173         /* Reset potential sums */
 174         velecsum         = _mm_setzero_ps();
 175         vvdwsum          = _mm_setzero_ps();
 176
 177         /* Start inner kernel loop */
 178         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 179         {
 180
 181             /* Get j neighbor index, and coordinate index */
 182             jnrA             = jjnr[jidx];
 183             jnrB             = jjnr[jidx+1];
 184             jnrC             = jjnr[jidx+2];
 185             jnrD             = jjnr[jidx+3];
 186             j_coord_offsetA  = DIM*jnrA;
 187             j_coord_offsetB  = DIM*jnrB;
 188             j_coord_offsetC  = DIM*jnrC;
 189             j_coord_offsetD  = DIM*jnrD;
 190
 191             /* load j atom coordinates */
 192             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 193                                               x+j_coord_offsetC,x+j_coord_offsetD,
 194                                               &jx0,&jy0,&jz0);
 195
 196             /* Calculate displacement vector */
 197             dx00             = _mm_sub_ps(ix0,jx0);
 198             dy00             = _mm_sub_ps(iy0,jy0);
 199             dz00             = _mm_sub_ps(iz0,jz0);
 200             dx10             = _mm_sub_ps(ix1,jx0);
 201             dy10             = _mm_sub_ps(iy1,jy0);
 202             dz10             = _mm_sub_ps(iz1,jz0);
 203             dx20             = _mm_sub_ps(ix2,jx0);
 204             dy20             = _mm_sub_ps(iy2,jy0);
 205             dz20             = _mm_sub_ps(iz2,jz0);
 206
 207             /* Calculate squared distance and things based on it */
 208             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 209             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 210             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 211
 212             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 213             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 214             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 215
 216             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 217             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 218             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 219
 220             /* Load parameters for j particles */
 221             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 222                                                               charge+jnrC+0,charge+jnrD+0);
 223             vdwjidx0A        = 2*vdwtype[jnrA+0];
 224             vdwjidx0B        = 2*vdwtype[jnrB+0];
 225             vdwjidx0C        = 2*vdwtype[jnrC+0];
 226             vdwjidx0D        = 2*vdwtype[jnrD+0];
 227
 228             fjx0             = _mm_setzero_ps();
 229             fjy0             = _mm_setzero_ps();
 230             fjz0             = _mm_setzero_ps();
 231
 232             /**************************
 233              * CALCULATE INTERACTIONS *
 234              **************************/
 235
 236             /* Compute parameters for interactions between i and j atoms */
 237             qq00             = _mm_mul_ps(iq0,jq0);
 238             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 239                                          vdwparam+vdwioffset0+vdwjidx0B,
 240                                          vdwparam+vdwioffset0+vdwjidx0C,
 241                                          vdwparam+vdwioffset0+vdwjidx0D,
 242                                          &c6_00,&c12_00);
 243
 244             /* COULOMB ELECTROSTATICS */
 245             velec            = _mm_mul_ps(qq00,rinv00);
 246             felec            = _mm_mul_ps(velec,rinvsq00);
 247
 248             /* LENNARD-JONES DISPERSION/REPULSION */
 249
 250             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 251             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 252             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 253             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 254             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 255
 256             /* Update potential sum for this i atom from the interaction with this j atom. */
 257             velecsum         = _mm_add_ps(velecsum,velec);
 258             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 259
 260             fscal            = _mm_add_ps(felec,fvdw);
 261
 262             /* Calculate temporary vectorial force */
 263             tx               = _mm_mul_ps(fscal,dx00);
 264             ty               = _mm_mul_ps(fscal,dy00);
 265             tz               = _mm_mul_ps(fscal,dz00);
 266
 267             /* Update vectorial force */
 268             fix0             = _mm_add_ps(fix0,tx);
 269             fiy0             = _mm_add_ps(fiy0,ty);
 270             fiz0             = _mm_add_ps(fiz0,tz);
 271
 272             fjx0             = _mm_add_ps(fjx0,tx);
 273             fjy0             = _mm_add_ps(fjy0,ty);
 274             fjz0             = _mm_add_ps(fjz0,tz);
 275
 276             /**************************
 277              * CALCULATE INTERACTIONS *
 278              **************************/
 279
 280             /* Compute parameters for interactions between i and j atoms */
 281             qq10             = _mm_mul_ps(iq1,jq0);
 282
 283             /* COULOMB ELECTROSTATICS */
 284             velec            = _mm_mul_ps(qq10,rinv10);
 285             felec            = _mm_mul_ps(velec,rinvsq10);
 286
 287             /* Update potential sum for this i atom from the interaction with this j atom. */
 288             velecsum         = _mm_add_ps(velecsum,velec);
 289
 290             fscal            = felec;
 291
 292             /* Calculate temporary vectorial force */
 293             tx               = _mm_mul_ps(fscal,dx10);
 294             ty               = _mm_mul_ps(fscal,dy10);
 295             tz               = _mm_mul_ps(fscal,dz10);
 296
 297             /* Update vectorial force */
 298             fix1             = _mm_add_ps(fix1,tx);
 299             fiy1             = _mm_add_ps(fiy1,ty);
 300             fiz1             = _mm_add_ps(fiz1,tz);
 301
 302             fjx0             = _mm_add_ps(fjx0,tx);
 303             fjy0             = _mm_add_ps(fjy0,ty);
 304             fjz0             = _mm_add_ps(fjz0,tz);
 305
 306             /**************************
 307              * CALCULATE INTERACTIONS *
 308              **************************/
 309
 310             /* Compute parameters for interactions between i and j atoms */
 311             qq20             = _mm_mul_ps(iq2,jq0);
 312
 313             /* COULOMB ELECTROSTATICS */
 314             velec            = _mm_mul_ps(qq20,rinv20);
 315             felec            = _mm_mul_ps(velec,rinvsq20);
 316
 317             /* Update potential sum for this i atom from the interaction with this j atom. */
 318             velecsum         = _mm_add_ps(velecsum,velec);
 319
 320             fscal            = felec;
 321
 322             /* Calculate temporary vectorial force */
 323             tx               = _mm_mul_ps(fscal,dx20);
 324             ty               = _mm_mul_ps(fscal,dy20);
 325             tz               = _mm_mul_ps(fscal,dz20);
 326
 327             /* Update vectorial force */
 328             fix2             = _mm_add_ps(fix2,tx);
 329             fiy2             = _mm_add_ps(fiy2,ty);
 330             fiz2             = _mm_add_ps(fiz2,tz);
 331
 332             fjx0             = _mm_add_ps(fjx0,tx);
 333             fjy0             = _mm_add_ps(fjy0,ty);
 334             fjz0             = _mm_add_ps(fjz0,tz);
 335
 336             fjptrA             = f+j_coord_offsetA;
 337             fjptrB             = f+j_coord_offsetB;
 338             fjptrC             = f+j_coord_offsetC;
 339             fjptrD             = f+j_coord_offsetD;
 340
 341             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 342
 343             /* Inner loop uses 96 flops */
 344         }
 345
 346         if(jidx<j_index_end)
 347         {
 348
 349             /* Get j neighbor index, and coordinate index */
 350             jnrlistA         = jjnr[jidx];
 351             jnrlistB         = jjnr[jidx+1];
 352             jnrlistC         = jjnr[jidx+2];
 353             jnrlistD         = jjnr[jidx+3];
 354             /* Sign of each element will be negative for non-real atoms.
 355              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 356              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 357              */
 358             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 359             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 360             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 361             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 362             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 363             j_coord_offsetA  = DIM*jnrA;
 364             j_coord_offsetB  = DIM*jnrB;
 365             j_coord_offsetC  = DIM*jnrC;
 366             j_coord_offsetD  = DIM*jnrD;
 367
 368             /* load j atom coordinates */
 369             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 370                                               x+j_coord_offsetC,x+j_coord_offsetD,
 371                                               &jx0,&jy0,&jz0);
 372
 373             /* Calculate displacement vector */
 374             dx00             = _mm_sub_ps(ix0,jx0);
 375             dy00             = _mm_sub_ps(iy0,jy0);
 376             dz00             = _mm_sub_ps(iz0,jz0);
 377             dx10             = _mm_sub_ps(ix1,jx0);
 378             dy10             = _mm_sub_ps(iy1,jy0);
 379             dz10             = _mm_sub_ps(iz1,jz0);
 380             dx20             = _mm_sub_ps(ix2,jx0);
 381             dy20             = _mm_sub_ps(iy2,jy0);
 382             dz20             = _mm_sub_ps(iz2,jz0);
 383
 384             /* Calculate squared distance and things based on it */
 385             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 386             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 387             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 388
 389             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 390             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 391             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 392
 393             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 394             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 395             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 396
 397             /* Load parameters for j particles */
 398             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 399                                                               charge+jnrC+0,charge+jnrD+0);
 400             vdwjidx0A        = 2*vdwtype[jnrA+0];
 401             vdwjidx0B        = 2*vdwtype[jnrB+0];
 402             vdwjidx0C        = 2*vdwtype[jnrC+0];
 403             vdwjidx0D        = 2*vdwtype[jnrD+0];
 404
 405             fjx0             = _mm_setzero_ps();
 406             fjy0             = _mm_setzero_ps();
 407             fjz0             = _mm_setzero_ps();
 408
 409             /**************************
 410              * CALCULATE INTERACTIONS *
 411              **************************/
 412
 413             /* Compute parameters for interactions between i and j atoms */
 414             qq00             = _mm_mul_ps(iq0,jq0);
 415             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 416                                          vdwparam+vdwioffset0+vdwjidx0B,
 417                                          vdwparam+vdwioffset0+vdwjidx0C,
 418                                          vdwparam+vdwioffset0+vdwjidx0D,
 419                                          &c6_00,&c12_00);
 420
 421             /* COULOMB ELECTROSTATICS */
 422             velec            = _mm_mul_ps(qq00,rinv00);
 423             felec            = _mm_mul_ps(velec,rinvsq00);
 424
 425             /* LENNARD-JONES DISPERSION/REPULSION */
 426
 427             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 428             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 429             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 430             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 431             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 432
 433             /* Update potential sum for this i atom from the interaction with this j atom. */
 434             velec            = _mm_andnot_ps(dummy_mask,velec);
 435             velecsum         = _mm_add_ps(velecsum,velec);
 436             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 437             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 438
 439             fscal            = _mm_add_ps(felec,fvdw);
 440
 441             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 442
 443             /* Calculate temporary vectorial force */
 444             tx               = _mm_mul_ps(fscal,dx00);
 445             ty               = _mm_mul_ps(fscal,dy00);
 446             tz               = _mm_mul_ps(fscal,dz00);
 447
 448             /* Update vectorial force */
 449             fix0             = _mm_add_ps(fix0,tx);
 450             fiy0             = _mm_add_ps(fiy0,ty);
 451             fiz0             = _mm_add_ps(fiz0,tz);
 452
 453             fjx0             = _mm_add_ps(fjx0,tx);
 454             fjy0             = _mm_add_ps(fjy0,ty);
 455             fjz0             = _mm_add_ps(fjz0,tz);
 456
 457             /**************************
 458              * CALCULATE INTERACTIONS *
 459              **************************/
 460
 461             /* Compute parameters for interactions between i and j atoms */
 462             qq10             = _mm_mul_ps(iq1,jq0);
 463
 464             /* COULOMB ELECTROSTATICS */
 465             velec            = _mm_mul_ps(qq10,rinv10);
 466             felec            = _mm_mul_ps(velec,rinvsq10);
 467
 468             /* Update potential sum for this i atom from the interaction with this j atom. */
 469             velec            = _mm_andnot_ps(dummy_mask,velec);
 470             velecsum         = _mm_add_ps(velecsum,velec);
 471
 472             fscal            = felec;
 473
 474             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 475
 476             /* Calculate temporary vectorial force */
 477             tx               = _mm_mul_ps(fscal,dx10);
 478             ty               = _mm_mul_ps(fscal,dy10);
 479             tz               = _mm_mul_ps(fscal,dz10);
 480
 481             /* Update vectorial force */
 482             fix1             = _mm_add_ps(fix1,tx);
 483             fiy1             = _mm_add_ps(fiy1,ty);
 484             fiz1             = _mm_add_ps(fiz1,tz);
 485
 486             fjx0             = _mm_add_ps(fjx0,tx);
 487             fjy0             = _mm_add_ps(fjy0,ty);
 488             fjz0             = _mm_add_ps(fjz0,tz);
 489
 490             /**************************
 491              * CALCULATE INTERACTIONS *
 492              **************************/
 493
 494             /* Compute parameters for interactions between i and j atoms */
 495             qq20             = _mm_mul_ps(iq2,jq0);
 496
 497             /* COULOMB ELECTROSTATICS */
 498             velec            = _mm_mul_ps(qq20,rinv20);
 499             felec            = _mm_mul_ps(velec,rinvsq20);
 500
 501             /* Update potential sum for this i atom from the interaction with this j atom. */
 502             velec            = _mm_andnot_ps(dummy_mask,velec);
 503             velecsum         = _mm_add_ps(velecsum,velec);
 504
 505             fscal            = felec;
 506
 507             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 508
 509             /* Calculate temporary vectorial force */
 510             tx               = _mm_mul_ps(fscal,dx20);
 511             ty               = _mm_mul_ps(fscal,dy20);
 512             tz               = _mm_mul_ps(fscal,dz20);
 513
 514             /* Update vectorial force */
 515             fix2             = _mm_add_ps(fix2,tx);
 516             fiy2             = _mm_add_ps(fiy2,ty);
 517             fiz2             = _mm_add_ps(fiz2,tz);
 518
 519             fjx0             = _mm_add_ps(fjx0,tx);
 520             fjy0             = _mm_add_ps(fjy0,ty);
 521             fjz0             = _mm_add_ps(fjz0,tz);
 522
 523             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 524             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 525             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 526             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 527
 528             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 529
 530             /* Inner loop uses 96 flops */
 531         }
 532
 533         /* End of innermost loop */
 534
 535         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 536                                               f+i_coord_offset,fshift+i_shift_offset);
 537
 538         ggid                        = gid[iidx];
 539         /* Update potential energies */
 540         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 541         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 542
 543         /* Increment number of inner iterations */
 544         inneriter                  += j_index_end - j_index_start;
 545
 546         /* Outer loop uses 20 flops */
 547     }
 548
 549     /* Increment number of outer iterations */
 550     outeriter        += nri;
 551
 552     /* Update outer/inner flops */
 553
 554     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*96);
 555 }
 556 /*
 557  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_sse4_1_single
 558  * Electrostatics interaction: Coulomb
 559  * VdW interaction:            LennardJones
 560  * Geometry:                   Water3-Particle
 561  * Calculate force/pot:        Force
 562  */
 563 void
 564 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_sse4_1_single
 565                     (t_nblist                    * gmx_restrict       nlist,
 566                      rvec                        * gmx_restrict          xx,
 567                      rvec                        * gmx_restrict          ff,
 568                      t_forcerec                  * gmx_restrict          fr,
 569                      t_mdatoms                   * gmx_restrict     mdatoms,
 570                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 571                      t_nrnb                      * gmx_restrict        nrnb)
 572 {
 573     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 574      * just 0 for non-waters.
 575      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 576      * jnr indices corresponding to data put in the four positions in the SIMD register.
 577      */
 578     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 579     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 580     int              jnrA,jnrB,jnrC,jnrD;
 581     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 582     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 583     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 584     real             rcutoff_scalar;
 585     real             *shiftvec,*fshift,*x,*f;
 586     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 587     real             scratch[4*DIM];
 588     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 589     int              vdwioffset0;
 590     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 591     int              vdwioffset1;
 592     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 593     int              vdwioffset2;
 594     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 595     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 596     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 597     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 598     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 599     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 600     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 601     real             *charge;
 602     int              nvdwtype;
 603     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 604     int              *vdwtype;
 605     real             *vdwparam;
 606     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 607     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 608     __m128           dummy_mask,cutoff_mask;
 609     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 610     __m128           one     = _mm_set1_ps(1.0);
 611     __m128           two     = _mm_set1_ps(2.0);
 612     x                = xx[0];
 613     f                = ff[0];
 614
 615     nri              = nlist->nri;
 616     iinr             = nlist->iinr;
 617     jindex           = nlist->jindex;
 618     jjnr             = nlist->jjnr;
 619     shiftidx         = nlist->shift;
 620     gid              = nlist->gid;
 621     shiftvec         = fr->shift_vec[0];
 622     fshift           = fr->fshift[0];
 623     facel            = _mm_set1_ps(fr->epsfac);
 624     charge           = mdatoms->chargeA;
 625     nvdwtype         = fr->ntype;
 626     vdwparam         = fr->nbfp;
 627     vdwtype          = mdatoms->typeA;
 628
 629     /* Setup water-specific parameters */
 630     inr              = nlist->iinr[0];
 631     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 632     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 633     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 634     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 635
 636     /* Avoid stupid compiler warnings */
 637     jnrA = jnrB = jnrC = jnrD = 0;
 638     j_coord_offsetA = 0;
 639     j_coord_offsetB = 0;
 640     j_coord_offsetC = 0;
 641     j_coord_offsetD = 0;
 642
 643     outeriter        = 0;
 644     inneriter        = 0;
 645
 646     for(iidx=0;iidx<4*DIM;iidx++)
 647     {
 648         scratch[iidx] = 0.0;
 649     }
 650
 651     /* Start outer loop over neighborlists */
 652     for(iidx=0; iidx<nri; iidx++)
 653     {
 654         /* Load shift vector for this list */
 655         i_shift_offset   = DIM*shiftidx[iidx];
 656
 657         /* Load limits for loop over neighbors */
 658         j_index_start    = jindex[iidx];
 659         j_index_end      = jindex[iidx+1];
 660
 661         /* Get outer coordinate index */
 662         inr              = iinr[iidx];
 663         i_coord_offset   = DIM*inr;
 664
 665         /* Load i particle coords and add shift vector */
 666         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 667                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 668
 669         fix0             = _mm_setzero_ps();
 670         fiy0             = _mm_setzero_ps();
 671         fiz0             = _mm_setzero_ps();
 672         fix1             = _mm_setzero_ps();
 673         fiy1             = _mm_setzero_ps();
 674         fiz1             = _mm_setzero_ps();
 675         fix2             = _mm_setzero_ps();
 676         fiy2             = _mm_setzero_ps();
 677         fiz2             = _mm_setzero_ps();
 678
 679         /* Start inner kernel loop */
 680         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 681         {
 682
 683             /* Get j neighbor index, and coordinate index */
 684             jnrA             = jjnr[jidx];
 685             jnrB             = jjnr[jidx+1];
 686             jnrC             = jjnr[jidx+2];
 687             jnrD             = jjnr[jidx+3];
 688             j_coord_offsetA  = DIM*jnrA;
 689             j_coord_offsetB  = DIM*jnrB;
 690             j_coord_offsetC  = DIM*jnrC;
 691             j_coord_offsetD  = DIM*jnrD;
 692
 693             /* load j atom coordinates */
 694             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 695                                               x+j_coord_offsetC,x+j_coord_offsetD,
 696                                               &jx0,&jy0,&jz0);
 697
 698             /* Calculate displacement vector */
 699             dx00             = _mm_sub_ps(ix0,jx0);
 700             dy00             = _mm_sub_ps(iy0,jy0);
 701             dz00             = _mm_sub_ps(iz0,jz0);
 702             dx10             = _mm_sub_ps(ix1,jx0);
 703             dy10             = _mm_sub_ps(iy1,jy0);
 704             dz10             = _mm_sub_ps(iz1,jz0);
 705             dx20             = _mm_sub_ps(ix2,jx0);
 706             dy20             = _mm_sub_ps(iy2,jy0);
 707             dz20             = _mm_sub_ps(iz2,jz0);
 708
 709             /* Calculate squared distance and things based on it */
 710             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 711             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 712             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 713
 714             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 715             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 716             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 717
 718             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 719             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 720             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 721
 722             /* Load parameters for j particles */
 723             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 724                                                               charge+jnrC+0,charge+jnrD+0);
 725             vdwjidx0A        = 2*vdwtype[jnrA+0];
 726             vdwjidx0B        = 2*vdwtype[jnrB+0];
 727             vdwjidx0C        = 2*vdwtype[jnrC+0];
 728             vdwjidx0D        = 2*vdwtype[jnrD+0];
 729
 730             fjx0             = _mm_setzero_ps();
 731             fjy0             = _mm_setzero_ps();
 732             fjz0             = _mm_setzero_ps();
 733
 734             /**************************
 735              * CALCULATE INTERACTIONS *
 736              **************************/
 737
 738             /* Compute parameters for interactions between i and j atoms */
 739             qq00             = _mm_mul_ps(iq0,jq0);
 740             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 741                                          vdwparam+vdwioffset0+vdwjidx0B,
 742                                          vdwparam+vdwioffset0+vdwjidx0C,
 743                                          vdwparam+vdwioffset0+vdwjidx0D,
 744                                          &c6_00,&c12_00);
 745
 746             /* COULOMB ELECTROSTATICS */
 747             velec            = _mm_mul_ps(qq00,rinv00);
 748             felec            = _mm_mul_ps(velec,rinvsq00);
 749
 750             /* LENNARD-JONES DISPERSION/REPULSION */
 751
 752             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 753             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 754
 755             fscal            = _mm_add_ps(felec,fvdw);
 756
 757             /* Calculate temporary vectorial force */
 758             tx               = _mm_mul_ps(fscal,dx00);
 759             ty               = _mm_mul_ps(fscal,dy00);
 760             tz               = _mm_mul_ps(fscal,dz00);
 761
 762             /* Update vectorial force */
 763             fix0             = _mm_add_ps(fix0,tx);
 764             fiy0             = _mm_add_ps(fiy0,ty);
 765             fiz0             = _mm_add_ps(fiz0,tz);
 766
 767             fjx0             = _mm_add_ps(fjx0,tx);
 768             fjy0             = _mm_add_ps(fjy0,ty);
 769             fjz0             = _mm_add_ps(fjz0,tz);
 770
 771             /**************************
 772              * CALCULATE INTERACTIONS *
 773              **************************/
 774
 775             /* Compute parameters for interactions between i and j atoms */
 776             qq10             = _mm_mul_ps(iq1,jq0);
 777
 778             /* COULOMB ELECTROSTATICS */
 779             velec            = _mm_mul_ps(qq10,rinv10);
 780             felec            = _mm_mul_ps(velec,rinvsq10);
 781
 782             fscal            = felec;
 783
 784             /* Calculate temporary vectorial force */
 785             tx               = _mm_mul_ps(fscal,dx10);
 786             ty               = _mm_mul_ps(fscal,dy10);
 787             tz               = _mm_mul_ps(fscal,dz10);
 788
 789             /* Update vectorial force */
 790             fix1             = _mm_add_ps(fix1,tx);
 791             fiy1             = _mm_add_ps(fiy1,ty);
 792             fiz1             = _mm_add_ps(fiz1,tz);
 793
 794             fjx0             = _mm_add_ps(fjx0,tx);
 795             fjy0             = _mm_add_ps(fjy0,ty);
 796             fjz0             = _mm_add_ps(fjz0,tz);
 797
 798             /**************************
 799              * CALCULATE INTERACTIONS *
 800              **************************/
 801
 802             /* Compute parameters for interactions between i and j atoms */
 803             qq20             = _mm_mul_ps(iq2,jq0);
 804
 805             /* COULOMB ELECTROSTATICS */
 806             velec            = _mm_mul_ps(qq20,rinv20);
 807             felec            = _mm_mul_ps(velec,rinvsq20);
 808
 809             fscal            = felec;
 810
 811             /* Calculate temporary vectorial force */
 812             tx               = _mm_mul_ps(fscal,dx20);
 813             ty               = _mm_mul_ps(fscal,dy20);
 814             tz               = _mm_mul_ps(fscal,dz20);
 815
 816             /* Update vectorial force */
 817             fix2             = _mm_add_ps(fix2,tx);
 818             fiy2             = _mm_add_ps(fiy2,ty);
 819             fiz2             = _mm_add_ps(fiz2,tz);
 820
 821             fjx0             = _mm_add_ps(fjx0,tx);
 822             fjy0             = _mm_add_ps(fjy0,ty);
 823             fjz0             = _mm_add_ps(fjz0,tz);
 824
 825             fjptrA             = f+j_coord_offsetA;
 826             fjptrB             = f+j_coord_offsetB;
 827             fjptrC             = f+j_coord_offsetC;
 828             fjptrD             = f+j_coord_offsetD;
 829
 830             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 831
 832             /* Inner loop uses 88 flops */
 833         }
 834
 835         if(jidx<j_index_end)
 836         {
 837
 838             /* Get j neighbor index, and coordinate index */
 839             jnrlistA         = jjnr[jidx];
 840             jnrlistB         = jjnr[jidx+1];
 841             jnrlistC         = jjnr[jidx+2];
 842             jnrlistD         = jjnr[jidx+3];
 843             /* Sign of each element will be negative for non-real atoms.
 844              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 845              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 846              */
 847             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 848             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 849             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 850             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 851             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 852             j_coord_offsetA  = DIM*jnrA;
 853             j_coord_offsetB  = DIM*jnrB;
 854             j_coord_offsetC  = DIM*jnrC;
 855             j_coord_offsetD  = DIM*jnrD;
 856
 857             /* load j atom coordinates */
 858             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 859                                               x+j_coord_offsetC,x+j_coord_offsetD,
 860                                               &jx0,&jy0,&jz0);
 861
 862             /* Calculate displacement vector */
 863             dx00             = _mm_sub_ps(ix0,jx0);
 864             dy00             = _mm_sub_ps(iy0,jy0);
 865             dz00             = _mm_sub_ps(iz0,jz0);
 866             dx10             = _mm_sub_ps(ix1,jx0);
 867             dy10             = _mm_sub_ps(iy1,jy0);
 868             dz10             = _mm_sub_ps(iz1,jz0);
 869             dx20             = _mm_sub_ps(ix2,jx0);
 870             dy20             = _mm_sub_ps(iy2,jy0);
 871             dz20             = _mm_sub_ps(iz2,jz0);
 872
 873             /* Calculate squared distance and things based on it */
 874             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 875             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 876             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 877
 878             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 879             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 880             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 881
 882             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 883             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 884             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 885
 886             /* Load parameters for j particles */
 887             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 888                                                               charge+jnrC+0,charge+jnrD+0);
 889             vdwjidx0A        = 2*vdwtype[jnrA+0];
 890             vdwjidx0B        = 2*vdwtype[jnrB+0];
 891             vdwjidx0C        = 2*vdwtype[jnrC+0];
 892             vdwjidx0D        = 2*vdwtype[jnrD+0];
 893
 894             fjx0             = _mm_setzero_ps();
 895             fjy0             = _mm_setzero_ps();
 896             fjz0             = _mm_setzero_ps();
 897
 898             /**************************
 899              * CALCULATE INTERACTIONS *
 900              **************************/
 901
 902             /* Compute parameters for interactions between i and j atoms */
 903             qq00             = _mm_mul_ps(iq0,jq0);
 904             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 905                                          vdwparam+vdwioffset0+vdwjidx0B,
 906                                          vdwparam+vdwioffset0+vdwjidx0C,
 907                                          vdwparam+vdwioffset0+vdwjidx0D,
 908                                          &c6_00,&c12_00);
 909
 910             /* COULOMB ELECTROSTATICS */
 911             velec            = _mm_mul_ps(qq00,rinv00);
 912             felec            = _mm_mul_ps(velec,rinvsq00);
 913
 914             /* LENNARD-JONES DISPERSION/REPULSION */
 915
 916             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 917             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 918
 919             fscal            = _mm_add_ps(felec,fvdw);
 920
 921             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 922
 923             /* Calculate temporary vectorial force */
 924             tx               = _mm_mul_ps(fscal,dx00);
 925             ty               = _mm_mul_ps(fscal,dy00);
 926             tz               = _mm_mul_ps(fscal,dz00);
 927
 928             /* Update vectorial force */
 929             fix0             = _mm_add_ps(fix0,tx);
 930             fiy0             = _mm_add_ps(fiy0,ty);
 931             fiz0             = _mm_add_ps(fiz0,tz);
 932
 933             fjx0             = _mm_add_ps(fjx0,tx);
 934             fjy0             = _mm_add_ps(fjy0,ty);
 935             fjz0             = _mm_add_ps(fjz0,tz);
 936
 937             /**************************
 938              * CALCULATE INTERACTIONS *
 939              **************************/
 940
 941             /* Compute parameters for interactions between i and j atoms */
 942             qq10             = _mm_mul_ps(iq1,jq0);
 943
 944             /* COULOMB ELECTROSTATICS */
 945             velec            = _mm_mul_ps(qq10,rinv10);
 946             felec            = _mm_mul_ps(velec,rinvsq10);
 947
 948             fscal            = felec;
 949
 950             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 951
 952             /* Calculate temporary vectorial force */
 953             tx               = _mm_mul_ps(fscal,dx10);
 954             ty               = _mm_mul_ps(fscal,dy10);
 955             tz               = _mm_mul_ps(fscal,dz10);
 956
 957             /* Update vectorial force */
 958             fix1             = _mm_add_ps(fix1,tx);
 959             fiy1             = _mm_add_ps(fiy1,ty);
 960             fiz1             = _mm_add_ps(fiz1,tz);
 961
 962             fjx0             = _mm_add_ps(fjx0,tx);
 963             fjy0             = _mm_add_ps(fjy0,ty);
 964             fjz0             = _mm_add_ps(fjz0,tz);
 965
 966             /**************************
 967              * CALCULATE INTERACTIONS *
 968              **************************/
 969
 970             /* Compute parameters for interactions between i and j atoms */
 971             qq20             = _mm_mul_ps(iq2,jq0);
 972
 973             /* COULOMB ELECTROSTATICS */
 974             velec            = _mm_mul_ps(qq20,rinv20);
 975             felec            = _mm_mul_ps(velec,rinvsq20);
 976
 977             fscal            = felec;
 978
 979             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 980
 981             /* Calculate temporary vectorial force */
 982             tx               = _mm_mul_ps(fscal,dx20);
 983             ty               = _mm_mul_ps(fscal,dy20);
 984             tz               = _mm_mul_ps(fscal,dz20);
 985
 986             /* Update vectorial force */
 987             fix2             = _mm_add_ps(fix2,tx);
 988             fiy2             = _mm_add_ps(fiy2,ty);
 989             fiz2             = _mm_add_ps(fiz2,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             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 996             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 997             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 998             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 999
1000             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1001
1002             /* Inner loop uses 88 flops */
1003         }
1004
1005         /* End of innermost loop */
1006
1007         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1008                                               f+i_coord_offset,fshift+i_shift_offset);
1009
1010         /* Increment number of inner iterations */
1011         inneriter                  += j_index_end - j_index_start;
1012
1013         /* Outer loop uses 18 flops */
1014     }
1015
1016     /* Increment number of outer iterations */
1017     outeriter        += nri;
1018
1019     /* Update outer/inner flops */
1020
1021     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*88);
1022 }