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

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse4_1_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW3P1_VF_sse4_1_single
  51  * Electrostatics interaction: Coulomb
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_VF_sse4_1_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  89     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 100     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 101     __m128           dummy_mask,cutoff_mask;
 102     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 103     __m128           one     = _mm_set1_ps(1.0);
 104     __m128           two     = _mm_set1_ps(2.0);
 105     x                = xx[0];
 106     f                = ff[0];
 107
 108     nri              = nlist->nri;
 109     iinr             = nlist->iinr;
 110     jindex           = nlist->jindex;
 111     jjnr             = nlist->jjnr;
 112     shiftidx         = nlist->shift;
 113     gid              = nlist->gid;
 114     shiftvec         = fr->shift_vec[0];
 115     fshift           = fr->fshift[0];
 116     facel            = _mm_set1_ps(fr->ic->epsfac);
 117     charge           = mdatoms->chargeA;
 118     nvdwtype         = fr->ntype;
 119     vdwparam         = fr->nbfp;
 120     vdwtype          = mdatoms->typeA;
 121
 122     /* Setup water-specific parameters */
 123     inr              = nlist->iinr[0];
 124     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 125     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 126     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 127     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 128
 129     /* Avoid stupid compiler warnings */
 130     jnrA = jnrB = jnrC = jnrD = 0;
 131     j_coord_offsetA = 0;
 132     j_coord_offsetB = 0;
 133     j_coord_offsetC = 0;
 134     j_coord_offsetD = 0;
 135
 136     outeriter        = 0;
 137     inneriter        = 0;
 138
 139     for(iidx=0;iidx<4*DIM;iidx++)
 140     {
 141         scratch[iidx] = 0.0;
 142     }
 143
 144     /* Start outer loop over neighborlists */
 145     for(iidx=0; iidx<nri; iidx++)
 146     {
 147         /* Load shift vector for this list */
 148         i_shift_offset   = DIM*shiftidx[iidx];
 149
 150         /* Load limits for loop over neighbors */
 151         j_index_start    = jindex[iidx];
 152         j_index_end      = jindex[iidx+1];
 153
 154         /* Get outer coordinate index */
 155         inr              = iinr[iidx];
 156         i_coord_offset   = DIM*inr;
 157
 158         /* Load i particle coords and add shift vector */
 159         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 160                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 161
 162         fix0             = _mm_setzero_ps();
 163         fiy0             = _mm_setzero_ps();
 164         fiz0             = _mm_setzero_ps();
 165         fix1             = _mm_setzero_ps();
 166         fiy1             = _mm_setzero_ps();
 167         fiz1             = _mm_setzero_ps();
 168         fix2             = _mm_setzero_ps();
 169         fiy2             = _mm_setzero_ps();
 170         fiz2             = _mm_setzero_ps();
 171
 172         /* Reset potential sums */
 173         velecsum         = _mm_setzero_ps();
 174         vvdwsum          = _mm_setzero_ps();
 175
 176         /* Start inner kernel loop */
 177         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 178         {
 179
 180             /* Get j neighbor index, and coordinate index */
 181             jnrA             = jjnr[jidx];
 182             jnrB             = jjnr[jidx+1];
 183             jnrC             = jjnr[jidx+2];
 184             jnrD             = jjnr[jidx+3];
 185             j_coord_offsetA  = DIM*jnrA;
 186             j_coord_offsetB  = DIM*jnrB;
 187             j_coord_offsetC  = DIM*jnrC;
 188             j_coord_offsetD  = DIM*jnrD;
 189
 190             /* load j atom coordinates */
 191             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 192                                               x+j_coord_offsetC,x+j_coord_offsetD,
 193                                               &jx0,&jy0,&jz0);
 194
 195             /* Calculate displacement vector */
 196             dx00             = _mm_sub_ps(ix0,jx0);
 197             dy00             = _mm_sub_ps(iy0,jy0);
 198             dz00             = _mm_sub_ps(iz0,jz0);
 199             dx10             = _mm_sub_ps(ix1,jx0);
 200             dy10             = _mm_sub_ps(iy1,jy0);
 201             dz10             = _mm_sub_ps(iz1,jz0);
 202             dx20             = _mm_sub_ps(ix2,jx0);
 203             dy20             = _mm_sub_ps(iy2,jy0);
 204             dz20             = _mm_sub_ps(iz2,jz0);
 205
 206             /* Calculate squared distance and things based on it */
 207             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 208             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 209             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 210
 211             rinv00           = sse41_invsqrt_f(rsq00);
 212             rinv10           = sse41_invsqrt_f(rsq10);
 213             rinv20           = sse41_invsqrt_f(rsq20);
 214
 215             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 216             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 217             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 218
 219             /* Load parameters for j particles */
 220             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 221                                                               charge+jnrC+0,charge+jnrD+0);
 222             vdwjidx0A        = 2*vdwtype[jnrA+0];
 223             vdwjidx0B        = 2*vdwtype[jnrB+0];
 224             vdwjidx0C        = 2*vdwtype[jnrC+0];
 225             vdwjidx0D        = 2*vdwtype[jnrD+0];
 226
 227             fjx0             = _mm_setzero_ps();
 228             fjy0             = _mm_setzero_ps();
 229             fjz0             = _mm_setzero_ps();
 230
 231             /**************************
 232              * CALCULATE INTERACTIONS *
 233              **************************/
 234
 235             /* Compute parameters for interactions between i and j atoms */
 236             qq00             = _mm_mul_ps(iq0,jq0);
 237             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 238                                          vdwparam+vdwioffset0+vdwjidx0B,
 239                                          vdwparam+vdwioffset0+vdwjidx0C,
 240                                          vdwparam+vdwioffset0+vdwjidx0D,
 241                                          &c6_00,&c12_00);
 242
 243             /* COULOMB ELECTROSTATICS */
 244             velec            = _mm_mul_ps(qq00,rinv00);
 245             felec            = _mm_mul_ps(velec,rinvsq00);
 246
 247             /* LENNARD-JONES DISPERSION/REPULSION */
 248
 249             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 250             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 251             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 252             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 253             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 254
 255             /* Update potential sum for this i atom from the interaction with this j atom. */
 256             velecsum         = _mm_add_ps(velecsum,velec);
 257             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 258
 259             fscal            = _mm_add_ps(felec,fvdw);
 260
 261             /* Calculate temporary vectorial force */
 262             tx               = _mm_mul_ps(fscal,dx00);
 263             ty               = _mm_mul_ps(fscal,dy00);
 264             tz               = _mm_mul_ps(fscal,dz00);
 265
 266             /* Update vectorial force */
 267             fix0             = _mm_add_ps(fix0,tx);
 268             fiy0             = _mm_add_ps(fiy0,ty);
 269             fiz0             = _mm_add_ps(fiz0,tz);
 270
 271             fjx0             = _mm_add_ps(fjx0,tx);
 272             fjy0             = _mm_add_ps(fjy0,ty);
 273             fjz0             = _mm_add_ps(fjz0,tz);
 274
 275             /**************************
 276              * CALCULATE INTERACTIONS *
 277              **************************/
 278
 279             /* Compute parameters for interactions between i and j atoms */
 280             qq10             = _mm_mul_ps(iq1,jq0);
 281
 282             /* COULOMB ELECTROSTATICS */
 283             velec            = _mm_mul_ps(qq10,rinv10);
 284             felec            = _mm_mul_ps(velec,rinvsq10);
 285
 286             /* Update potential sum for this i atom from the interaction with this j atom. */
 287             velecsum         = _mm_add_ps(velecsum,velec);
 288
 289             fscal            = felec;
 290
 291             /* Calculate temporary vectorial force */
 292             tx               = _mm_mul_ps(fscal,dx10);
 293             ty               = _mm_mul_ps(fscal,dy10);
 294             tz               = _mm_mul_ps(fscal,dz10);
 295
 296             /* Update vectorial force */
 297             fix1             = _mm_add_ps(fix1,tx);
 298             fiy1             = _mm_add_ps(fiy1,ty);
 299             fiz1             = _mm_add_ps(fiz1,tz);
 300
 301             fjx0             = _mm_add_ps(fjx0,tx);
 302             fjy0             = _mm_add_ps(fjy0,ty);
 303             fjz0             = _mm_add_ps(fjz0,tz);
 304
 305             /**************************
 306              * CALCULATE INTERACTIONS *
 307              **************************/
 308
 309             /* Compute parameters for interactions between i and j atoms */
 310             qq20             = _mm_mul_ps(iq2,jq0);
 311
 312             /* COULOMB ELECTROSTATICS */
 313             velec            = _mm_mul_ps(qq20,rinv20);
 314             felec            = _mm_mul_ps(velec,rinvsq20);
 315
 316             /* Update potential sum for this i atom from the interaction with this j atom. */
 317             velecsum         = _mm_add_ps(velecsum,velec);
 318
 319             fscal            = felec;
 320
 321             /* Calculate temporary vectorial force */
 322             tx               = _mm_mul_ps(fscal,dx20);
 323             ty               = _mm_mul_ps(fscal,dy20);
 324             tz               = _mm_mul_ps(fscal,dz20);
 325
 326             /* Update vectorial force */
 327             fix2             = _mm_add_ps(fix2,tx);
 328             fiy2             = _mm_add_ps(fiy2,ty);
 329             fiz2             = _mm_add_ps(fiz2,tz);
 330
 331             fjx0             = _mm_add_ps(fjx0,tx);
 332             fjy0             = _mm_add_ps(fjy0,ty);
 333             fjz0             = _mm_add_ps(fjz0,tz);
 334
 335             fjptrA             = f+j_coord_offsetA;
 336             fjptrB             = f+j_coord_offsetB;
 337             fjptrC             = f+j_coord_offsetC;
 338             fjptrD             = f+j_coord_offsetD;
 339
 340             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 341
 342             /* Inner loop uses 96 flops */
 343         }
 344
 345         if(jidx<j_index_end)
 346         {
 347
 348             /* Get j neighbor index, and coordinate index */
 349             jnrlistA         = jjnr[jidx];
 350             jnrlistB         = jjnr[jidx+1];
 351             jnrlistC         = jjnr[jidx+2];
 352             jnrlistD         = jjnr[jidx+3];
 353             /* Sign of each element will be negative for non-real atoms.
 354              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 355              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 356              */
 357             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 358             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 359             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 360             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 361             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 362             j_coord_offsetA  = DIM*jnrA;
 363             j_coord_offsetB  = DIM*jnrB;
 364             j_coord_offsetC  = DIM*jnrC;
 365             j_coord_offsetD  = DIM*jnrD;
 366
 367             /* load j atom coordinates */
 368             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 369                                               x+j_coord_offsetC,x+j_coord_offsetD,
 370                                               &jx0,&jy0,&jz0);
 371
 372             /* Calculate displacement vector */
 373             dx00             = _mm_sub_ps(ix0,jx0);
 374             dy00             = _mm_sub_ps(iy0,jy0);
 375             dz00             = _mm_sub_ps(iz0,jz0);
 376             dx10             = _mm_sub_ps(ix1,jx0);
 377             dy10             = _mm_sub_ps(iy1,jy0);
 378             dz10             = _mm_sub_ps(iz1,jz0);
 379             dx20             = _mm_sub_ps(ix2,jx0);
 380             dy20             = _mm_sub_ps(iy2,jy0);
 381             dz20             = _mm_sub_ps(iz2,jz0);
 382
 383             /* Calculate squared distance and things based on it */
 384             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 385             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 386             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 387
 388             rinv00           = sse41_invsqrt_f(rsq00);
 389             rinv10           = sse41_invsqrt_f(rsq10);
 390             rinv20           = sse41_invsqrt_f(rsq20);
 391
 392             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 393             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 394             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 395
 396             /* Load parameters for j particles */
 397             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 398                                                               charge+jnrC+0,charge+jnrD+0);
 399             vdwjidx0A        = 2*vdwtype[jnrA+0];
 400             vdwjidx0B        = 2*vdwtype[jnrB+0];
 401             vdwjidx0C        = 2*vdwtype[jnrC+0];
 402             vdwjidx0D        = 2*vdwtype[jnrD+0];
 403
 404             fjx0             = _mm_setzero_ps();
 405             fjy0             = _mm_setzero_ps();
 406             fjz0             = _mm_setzero_ps();
 407
 408             /**************************
 409              * CALCULATE INTERACTIONS *
 410              **************************/
 411
 412             /* Compute parameters for interactions between i and j atoms */
 413             qq00             = _mm_mul_ps(iq0,jq0);
 414             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 415                                          vdwparam+vdwioffset0+vdwjidx0B,
 416                                          vdwparam+vdwioffset0+vdwjidx0C,
 417                                          vdwparam+vdwioffset0+vdwjidx0D,
 418                                          &c6_00,&c12_00);
 419
 420             /* COULOMB ELECTROSTATICS */
 421             velec            = _mm_mul_ps(qq00,rinv00);
 422             felec            = _mm_mul_ps(velec,rinvsq00);
 423
 424             /* LENNARD-JONES DISPERSION/REPULSION */
 425
 426             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 427             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 428             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 429             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 430             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 431
 432             /* Update potential sum for this i atom from the interaction with this j atom. */
 433             velec            = _mm_andnot_ps(dummy_mask,velec);
 434             velecsum         = _mm_add_ps(velecsum,velec);
 435             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 436             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 437
 438             fscal            = _mm_add_ps(felec,fvdw);
 439
 440             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 441
 442             /* Calculate temporary vectorial force */
 443             tx               = _mm_mul_ps(fscal,dx00);
 444             ty               = _mm_mul_ps(fscal,dy00);
 445             tz               = _mm_mul_ps(fscal,dz00);
 446
 447             /* Update vectorial force */
 448             fix0             = _mm_add_ps(fix0,tx);
 449             fiy0             = _mm_add_ps(fiy0,ty);
 450             fiz0             = _mm_add_ps(fiz0,tz);
 451
 452             fjx0             = _mm_add_ps(fjx0,tx);
 453             fjy0             = _mm_add_ps(fjy0,ty);
 454             fjz0             = _mm_add_ps(fjz0,tz);
 455
 456             /**************************
 457              * CALCULATE INTERACTIONS *
 458              **************************/
 459
 460             /* Compute parameters for interactions between i and j atoms */
 461             qq10             = _mm_mul_ps(iq1,jq0);
 462
 463             /* COULOMB ELECTROSTATICS */
 464             velec            = _mm_mul_ps(qq10,rinv10);
 465             felec            = _mm_mul_ps(velec,rinvsq10);
 466
 467             /* Update potential sum for this i atom from the interaction with this j atom. */
 468             velec            = _mm_andnot_ps(dummy_mask,velec);
 469             velecsum         = _mm_add_ps(velecsum,velec);
 470
 471             fscal            = felec;
 472
 473             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 474
 475             /* Calculate temporary vectorial force */
 476             tx               = _mm_mul_ps(fscal,dx10);
 477             ty               = _mm_mul_ps(fscal,dy10);
 478             tz               = _mm_mul_ps(fscal,dz10);
 479
 480             /* Update vectorial force */
 481             fix1             = _mm_add_ps(fix1,tx);
 482             fiy1             = _mm_add_ps(fiy1,ty);
 483             fiz1             = _mm_add_ps(fiz1,tz);
 484
 485             fjx0             = _mm_add_ps(fjx0,tx);
 486             fjy0             = _mm_add_ps(fjy0,ty);
 487             fjz0             = _mm_add_ps(fjz0,tz);
 488
 489             /**************************
 490              * CALCULATE INTERACTIONS *
 491              **************************/
 492
 493             /* Compute parameters for interactions between i and j atoms */
 494             qq20             = _mm_mul_ps(iq2,jq0);
 495
 496             /* COULOMB ELECTROSTATICS */
 497             velec            = _mm_mul_ps(qq20,rinv20);
 498             felec            = _mm_mul_ps(velec,rinvsq20);
 499
 500             /* Update potential sum for this i atom from the interaction with this j atom. */
 501             velec            = _mm_andnot_ps(dummy_mask,velec);
 502             velecsum         = _mm_add_ps(velecsum,velec);
 503
 504             fscal            = felec;
 505
 506             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 507
 508             /* Calculate temporary vectorial force */
 509             tx               = _mm_mul_ps(fscal,dx20);
 510             ty               = _mm_mul_ps(fscal,dy20);
 511             tz               = _mm_mul_ps(fscal,dz20);
 512
 513             /* Update vectorial force */
 514             fix2             = _mm_add_ps(fix2,tx);
 515             fiy2             = _mm_add_ps(fiy2,ty);
 516             fiz2             = _mm_add_ps(fiz2,tz);
 517
 518             fjx0             = _mm_add_ps(fjx0,tx);
 519             fjy0             = _mm_add_ps(fjy0,ty);
 520             fjz0             = _mm_add_ps(fjz0,tz);
 521
 522             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 523             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 524             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 525             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 526
 527             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 528
 529             /* Inner loop uses 96 flops */
 530         }
 531
 532         /* End of innermost loop */
 533
 534         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 535                                               f+i_coord_offset,fshift+i_shift_offset);
 536
 537         ggid                        = gid[iidx];
 538         /* Update potential energies */
 539         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 540         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 541
 542         /* Increment number of inner iterations */
 543         inneriter                  += j_index_end - j_index_start;
 544
 545         /* Outer loop uses 20 flops */
 546     }
 547
 548     /* Increment number of outer iterations */
 549     outeriter        += nri;
 550
 551     /* Update outer/inner flops */
 552
 553     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*96);
 554 }
 555 /*
 556  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_sse4_1_single
 557  * Electrostatics interaction: Coulomb
 558  * VdW interaction:            LennardJones
 559  * Geometry:                   Water3-Particle
 560  * Calculate force/pot:        Force
 561  */
 562 void
 563 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_sse4_1_single
 564                     (t_nblist                    * gmx_restrict       nlist,
 565                      rvec                        * gmx_restrict          xx,
 566                      rvec                        * gmx_restrict          ff,
 567                      struct t_forcerec           * gmx_restrict          fr,
 568                      t_mdatoms                   * gmx_restrict     mdatoms,
 569                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 570                      t_nrnb                      * gmx_restrict        nrnb)
 571 {
 572     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 573      * just 0 for non-waters.
 574      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 575      * jnr indices corresponding to data put in the four positions in the SIMD register.
 576      */
 577     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 578     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 579     int              jnrA,jnrB,jnrC,jnrD;
 580     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 581     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 582     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 583     real             rcutoff_scalar;
 584     real             *shiftvec,*fshift,*x,*f;
 585     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 586     real             scratch[4*DIM];
 587     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 588     int              vdwioffset0;
 589     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 590     int              vdwioffset1;
 591     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 592     int              vdwioffset2;
 593     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 594     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 595     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 596     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 597     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 598     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 599     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 600     real             *charge;
 601     int              nvdwtype;
 602     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 603     int              *vdwtype;
 604     real             *vdwparam;
 605     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 606     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 607     __m128           dummy_mask,cutoff_mask;
 608     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 609     __m128           one     = _mm_set1_ps(1.0);
 610     __m128           two     = _mm_set1_ps(2.0);
 611     x                = xx[0];
 612     f                = ff[0];
 613
 614     nri              = nlist->nri;
 615     iinr             = nlist->iinr;
 616     jindex           = nlist->jindex;
 617     jjnr             = nlist->jjnr;
 618     shiftidx         = nlist->shift;
 619     gid              = nlist->gid;
 620     shiftvec         = fr->shift_vec[0];
 621     fshift           = fr->fshift[0];
 622     facel            = _mm_set1_ps(fr->ic->epsfac);
 623     charge           = mdatoms->chargeA;
 624     nvdwtype         = fr->ntype;
 625     vdwparam         = fr->nbfp;
 626     vdwtype          = mdatoms->typeA;
 627
 628     /* Setup water-specific parameters */
 629     inr              = nlist->iinr[0];
 630     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 631     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 632     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 633     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 634
 635     /* Avoid stupid compiler warnings */
 636     jnrA = jnrB = jnrC = jnrD = 0;
 637     j_coord_offsetA = 0;
 638     j_coord_offsetB = 0;
 639     j_coord_offsetC = 0;
 640     j_coord_offsetD = 0;
 641
 642     outeriter        = 0;
 643     inneriter        = 0;
 644
 645     for(iidx=0;iidx<4*DIM;iidx++)
 646     {
 647         scratch[iidx] = 0.0;
 648     }
 649
 650     /* Start outer loop over neighborlists */
 651     for(iidx=0; iidx<nri; iidx++)
 652     {
 653         /* Load shift vector for this list */
 654         i_shift_offset   = DIM*shiftidx[iidx];
 655
 656         /* Load limits for loop over neighbors */
 657         j_index_start    = jindex[iidx];
 658         j_index_end      = jindex[iidx+1];
 659
 660         /* Get outer coordinate index */
 661         inr              = iinr[iidx];
 662         i_coord_offset   = DIM*inr;
 663
 664         /* Load i particle coords and add shift vector */
 665         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 666                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 667
 668         fix0             = _mm_setzero_ps();
 669         fiy0             = _mm_setzero_ps();
 670         fiz0             = _mm_setzero_ps();
 671         fix1             = _mm_setzero_ps();
 672         fiy1             = _mm_setzero_ps();
 673         fiz1             = _mm_setzero_ps();
 674         fix2             = _mm_setzero_ps();
 675         fiy2             = _mm_setzero_ps();
 676         fiz2             = _mm_setzero_ps();
 677
 678         /* Start inner kernel loop */
 679         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 680         {
 681
 682             /* Get j neighbor index, and coordinate index */
 683             jnrA             = jjnr[jidx];
 684             jnrB             = jjnr[jidx+1];
 685             jnrC             = jjnr[jidx+2];
 686             jnrD             = jjnr[jidx+3];
 687             j_coord_offsetA  = DIM*jnrA;
 688             j_coord_offsetB  = DIM*jnrB;
 689             j_coord_offsetC  = DIM*jnrC;
 690             j_coord_offsetD  = DIM*jnrD;
 691
 692             /* load j atom coordinates */
 693             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 694                                               x+j_coord_offsetC,x+j_coord_offsetD,
 695                                               &jx0,&jy0,&jz0);
 696
 697             /* Calculate displacement vector */
 698             dx00             = _mm_sub_ps(ix0,jx0);
 699             dy00             = _mm_sub_ps(iy0,jy0);
 700             dz00             = _mm_sub_ps(iz0,jz0);
 701             dx10             = _mm_sub_ps(ix1,jx0);
 702             dy10             = _mm_sub_ps(iy1,jy0);
 703             dz10             = _mm_sub_ps(iz1,jz0);
 704             dx20             = _mm_sub_ps(ix2,jx0);
 705             dy20             = _mm_sub_ps(iy2,jy0);
 706             dz20             = _mm_sub_ps(iz2,jz0);
 707
 708             /* Calculate squared distance and things based on it */
 709             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 710             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 711             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 712
 713             rinv00           = sse41_invsqrt_f(rsq00);
 714             rinv10           = sse41_invsqrt_f(rsq10);
 715             rinv20           = sse41_invsqrt_f(rsq20);
 716
 717             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 718             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 719             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 720
 721             /* Load parameters for j particles */
 722             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 723                                                               charge+jnrC+0,charge+jnrD+0);
 724             vdwjidx0A        = 2*vdwtype[jnrA+0];
 725             vdwjidx0B        = 2*vdwtype[jnrB+0];
 726             vdwjidx0C        = 2*vdwtype[jnrC+0];
 727             vdwjidx0D        = 2*vdwtype[jnrD+0];
 728
 729             fjx0             = _mm_setzero_ps();
 730             fjy0             = _mm_setzero_ps();
 731             fjz0             = _mm_setzero_ps();
 732
 733             /**************************
 734              * CALCULATE INTERACTIONS *
 735              **************************/
 736
 737             /* Compute parameters for interactions between i and j atoms */
 738             qq00             = _mm_mul_ps(iq0,jq0);
 739             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 740                                          vdwparam+vdwioffset0+vdwjidx0B,
 741                                          vdwparam+vdwioffset0+vdwjidx0C,
 742                                          vdwparam+vdwioffset0+vdwjidx0D,
 743                                          &c6_00,&c12_00);
 744
 745             /* COULOMB ELECTROSTATICS */
 746             velec            = _mm_mul_ps(qq00,rinv00);
 747             felec            = _mm_mul_ps(velec,rinvsq00);
 748
 749             /* LENNARD-JONES DISPERSION/REPULSION */
 750
 751             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 752             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 753
 754             fscal            = _mm_add_ps(felec,fvdw);
 755
 756             /* Calculate temporary vectorial force */
 757             tx               = _mm_mul_ps(fscal,dx00);
 758             ty               = _mm_mul_ps(fscal,dy00);
 759             tz               = _mm_mul_ps(fscal,dz00);
 760
 761             /* Update vectorial force */
 762             fix0             = _mm_add_ps(fix0,tx);
 763             fiy0             = _mm_add_ps(fiy0,ty);
 764             fiz0             = _mm_add_ps(fiz0,tz);
 765
 766             fjx0             = _mm_add_ps(fjx0,tx);
 767             fjy0             = _mm_add_ps(fjy0,ty);
 768             fjz0             = _mm_add_ps(fjz0,tz);
 769
 770             /**************************
 771              * CALCULATE INTERACTIONS *
 772              **************************/
 773
 774             /* Compute parameters for interactions between i and j atoms */
 775             qq10             = _mm_mul_ps(iq1,jq0);
 776
 777             /* COULOMB ELECTROSTATICS */
 778             velec            = _mm_mul_ps(qq10,rinv10);
 779             felec            = _mm_mul_ps(velec,rinvsq10);
 780
 781             fscal            = felec;
 782
 783             /* Calculate temporary vectorial force */
 784             tx               = _mm_mul_ps(fscal,dx10);
 785             ty               = _mm_mul_ps(fscal,dy10);
 786             tz               = _mm_mul_ps(fscal,dz10);
 787
 788             /* Update vectorial force */
 789             fix1             = _mm_add_ps(fix1,tx);
 790             fiy1             = _mm_add_ps(fiy1,ty);
 791             fiz1             = _mm_add_ps(fiz1,tz);
 792
 793             fjx0             = _mm_add_ps(fjx0,tx);
 794             fjy0             = _mm_add_ps(fjy0,ty);
 795             fjz0             = _mm_add_ps(fjz0,tz);
 796
 797             /**************************
 798              * CALCULATE INTERACTIONS *
 799              **************************/
 800
 801             /* Compute parameters for interactions between i and j atoms */
 802             qq20             = _mm_mul_ps(iq2,jq0);
 803
 804             /* COULOMB ELECTROSTATICS */
 805             velec            = _mm_mul_ps(qq20,rinv20);
 806             felec            = _mm_mul_ps(velec,rinvsq20);
 807
 808             fscal            = felec;
 809
 810             /* Calculate temporary vectorial force */
 811             tx               = _mm_mul_ps(fscal,dx20);
 812             ty               = _mm_mul_ps(fscal,dy20);
 813             tz               = _mm_mul_ps(fscal,dz20);
 814
 815             /* Update vectorial force */
 816             fix2             = _mm_add_ps(fix2,tx);
 817             fiy2             = _mm_add_ps(fiy2,ty);
 818             fiz2             = _mm_add_ps(fiz2,tz);
 819
 820             fjx0             = _mm_add_ps(fjx0,tx);
 821             fjy0             = _mm_add_ps(fjy0,ty);
 822             fjz0             = _mm_add_ps(fjz0,tz);
 823
 824             fjptrA             = f+j_coord_offsetA;
 825             fjptrB             = f+j_coord_offsetB;
 826             fjptrC             = f+j_coord_offsetC;
 827             fjptrD             = f+j_coord_offsetD;
 828
 829             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 830
 831             /* Inner loop uses 88 flops */
 832         }
 833
 834         if(jidx<j_index_end)
 835         {
 836
 837             /* Get j neighbor index, and coordinate index */
 838             jnrlistA         = jjnr[jidx];
 839             jnrlistB         = jjnr[jidx+1];
 840             jnrlistC         = jjnr[jidx+2];
 841             jnrlistD         = jjnr[jidx+3];
 842             /* Sign of each element will be negative for non-real atoms.
 843              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 844              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 845              */
 846             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 847             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 848             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 849             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 850             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 851             j_coord_offsetA  = DIM*jnrA;
 852             j_coord_offsetB  = DIM*jnrB;
 853             j_coord_offsetC  = DIM*jnrC;
 854             j_coord_offsetD  = DIM*jnrD;
 855
 856             /* load j atom coordinates */
 857             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 858                                               x+j_coord_offsetC,x+j_coord_offsetD,
 859                                               &jx0,&jy0,&jz0);
 860
 861             /* Calculate displacement vector */
 862             dx00             = _mm_sub_ps(ix0,jx0);
 863             dy00             = _mm_sub_ps(iy0,jy0);
 864             dz00             = _mm_sub_ps(iz0,jz0);
 865             dx10             = _mm_sub_ps(ix1,jx0);
 866             dy10             = _mm_sub_ps(iy1,jy0);
 867             dz10             = _mm_sub_ps(iz1,jz0);
 868             dx20             = _mm_sub_ps(ix2,jx0);
 869             dy20             = _mm_sub_ps(iy2,jy0);
 870             dz20             = _mm_sub_ps(iz2,jz0);
 871
 872             /* Calculate squared distance and things based on it */
 873             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 874             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 875             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 876
 877             rinv00           = sse41_invsqrt_f(rsq00);
 878             rinv10           = sse41_invsqrt_f(rsq10);
 879             rinv20           = sse41_invsqrt_f(rsq20);
 880
 881             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 882             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 883             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 884
 885             /* Load parameters for j particles */
 886             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 887                                                               charge+jnrC+0,charge+jnrD+0);
 888             vdwjidx0A        = 2*vdwtype[jnrA+0];
 889             vdwjidx0B        = 2*vdwtype[jnrB+0];
 890             vdwjidx0C        = 2*vdwtype[jnrC+0];
 891             vdwjidx0D        = 2*vdwtype[jnrD+0];
 892
 893             fjx0             = _mm_setzero_ps();
 894             fjy0             = _mm_setzero_ps();
 895             fjz0             = _mm_setzero_ps();
 896
 897             /**************************
 898              * CALCULATE INTERACTIONS *
 899              **************************/
 900
 901             /* Compute parameters for interactions between i and j atoms */
 902             qq00             = _mm_mul_ps(iq0,jq0);
 903             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 904                                          vdwparam+vdwioffset0+vdwjidx0B,
 905                                          vdwparam+vdwioffset0+vdwjidx0C,
 906                                          vdwparam+vdwioffset0+vdwjidx0D,
 907                                          &c6_00,&c12_00);
 908
 909             /* COULOMB ELECTROSTATICS */
 910             velec            = _mm_mul_ps(qq00,rinv00);
 911             felec            = _mm_mul_ps(velec,rinvsq00);
 912
 913             /* LENNARD-JONES DISPERSION/REPULSION */
 914
 915             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 916             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 917
 918             fscal            = _mm_add_ps(felec,fvdw);
 919
 920             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 921
 922             /* Calculate temporary vectorial force */
 923             tx               = _mm_mul_ps(fscal,dx00);
 924             ty               = _mm_mul_ps(fscal,dy00);
 925             tz               = _mm_mul_ps(fscal,dz00);
 926
 927             /* Update vectorial force */
 928             fix0             = _mm_add_ps(fix0,tx);
 929             fiy0             = _mm_add_ps(fiy0,ty);
 930             fiz0             = _mm_add_ps(fiz0,tz);
 931
 932             fjx0             = _mm_add_ps(fjx0,tx);
 933             fjy0             = _mm_add_ps(fjy0,ty);
 934             fjz0             = _mm_add_ps(fjz0,tz);
 935
 936             /**************************
 937              * CALCULATE INTERACTIONS *
 938              **************************/
 939
 940             /* Compute parameters for interactions between i and j atoms */
 941             qq10             = _mm_mul_ps(iq1,jq0);
 942
 943             /* COULOMB ELECTROSTATICS */
 944             velec            = _mm_mul_ps(qq10,rinv10);
 945             felec            = _mm_mul_ps(velec,rinvsq10);
 946
 947             fscal            = felec;
 948
 949             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 950
 951             /* Calculate temporary vectorial force */
 952             tx               = _mm_mul_ps(fscal,dx10);
 953             ty               = _mm_mul_ps(fscal,dy10);
 954             tz               = _mm_mul_ps(fscal,dz10);
 955
 956             /* Update vectorial force */
 957             fix1             = _mm_add_ps(fix1,tx);
 958             fiy1             = _mm_add_ps(fiy1,ty);
 959             fiz1             = _mm_add_ps(fiz1,tz);
 960
 961             fjx0             = _mm_add_ps(fjx0,tx);
 962             fjy0             = _mm_add_ps(fjy0,ty);
 963             fjz0             = _mm_add_ps(fjz0,tz);
 964
 965             /**************************
 966              * CALCULATE INTERACTIONS *
 967              **************************/
 968
 969             /* Compute parameters for interactions between i and j atoms */
 970             qq20             = _mm_mul_ps(iq2,jq0);
 971
 972             /* COULOMB ELECTROSTATICS */
 973             velec            = _mm_mul_ps(qq20,rinv20);
 974             felec            = _mm_mul_ps(velec,rinvsq20);
 975
 976             fscal            = felec;
 977
 978             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 979
 980             /* Calculate temporary vectorial force */
 981             tx               = _mm_mul_ps(fscal,dx20);
 982             ty               = _mm_mul_ps(fscal,dy20);
 983             tz               = _mm_mul_ps(fscal,dz20);
 984
 985             /* Update vectorial force */
 986             fix2             = _mm_add_ps(fix2,tx);
 987             fiy2             = _mm_add_ps(fiy2,ty);
 988             fiz2             = _mm_add_ps(fiz2,tz);
 989
 990             fjx0             = _mm_add_ps(fjx0,tx);
 991             fjy0             = _mm_add_ps(fjy0,ty);
 992             fjz0             = _mm_add_ps(fjz0,tz);
 993
 994             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 995             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 996             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 997             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 998
 999             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1000
1001             /* Inner loop uses 88 flops */
1002         }
1003
1004         /* End of innermost loop */
1005
1006         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1007                                               f+i_coord_offset,fshift+i_shift_offset);
1008
1009         /* Increment number of inner iterations */
1010         inneriter                  += j_index_end - j_index_start;
1011
1012         /* Outer loop uses 18 flops */
1013     }
1014
1015     /* Increment number of outer iterations */
1016     outeriter        += nri;
1017
1018     /* Update outer/inner flops */
1019
1020     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*88);
1021 }