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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse4_1_single.h"
  50 #include "kernelutil_x86_sse4_1_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW3P1_VF_sse4_1_single
  54  * Electrostatics interaction: Coulomb
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_VF_sse4_1_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 103     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 104     __m128           dummy_mask,cutoff_mask;
 105     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 106     __m128           one     = _mm_set1_ps(1.0);
 107     __m128           two     = _mm_set1_ps(2.0);
 108     x                = xx[0];
 109     f                = ff[0];
 110
 111     nri              = nlist->nri;
 112     iinr             = nlist->iinr;
 113     jindex           = nlist->jindex;
 114     jjnr             = nlist->jjnr;
 115     shiftidx         = nlist->shift;
 116     gid              = nlist->gid;
 117     shiftvec         = fr->shift_vec[0];
 118     fshift           = fr->fshift[0];
 119     facel            = _mm_set1_ps(fr->epsfac);
 120     charge           = mdatoms->chargeA;
 121     nvdwtype         = fr->ntype;
 122     vdwparam         = fr->nbfp;
 123     vdwtype          = mdatoms->typeA;
 124
 125     /* Setup water-specific parameters */
 126     inr              = nlist->iinr[0];
 127     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 128     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 129     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 130     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 131
 132     /* Avoid stupid compiler warnings */
 133     jnrA = jnrB = jnrC = jnrD = 0;
 134     j_coord_offsetA = 0;
 135     j_coord_offsetB = 0;
 136     j_coord_offsetC = 0;
 137     j_coord_offsetD = 0;
 138
 139     outeriter        = 0;
 140     inneriter        = 0;
 141
 142     for(iidx=0;iidx<4*DIM;iidx++)
 143     {
 144         scratch[iidx] = 0.0;
 145     }
 146
 147     /* Start outer loop over neighborlists */
 148     for(iidx=0; iidx<nri; iidx++)
 149     {
 150         /* Load shift vector for this list */
 151         i_shift_offset   = DIM*shiftidx[iidx];
 152
 153         /* Load limits for loop over neighbors */
 154         j_index_start    = jindex[iidx];
 155         j_index_end      = jindex[iidx+1];
 156
 157         /* Get outer coordinate index */
 158         inr              = iinr[iidx];
 159         i_coord_offset   = DIM*inr;
 160
 161         /* Load i particle coords and add shift vector */
 162         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 163                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 164
 165         fix0             = _mm_setzero_ps();
 166         fiy0             = _mm_setzero_ps();
 167         fiz0             = _mm_setzero_ps();
 168         fix1             = _mm_setzero_ps();
 169         fiy1             = _mm_setzero_ps();
 170         fiz1             = _mm_setzero_ps();
 171         fix2             = _mm_setzero_ps();
 172         fiy2             = _mm_setzero_ps();
 173         fiz2             = _mm_setzero_ps();
 174
 175         /* Reset potential sums */
 176         velecsum         = _mm_setzero_ps();
 177         vvdwsum          = _mm_setzero_ps();
 178
 179         /* Start inner kernel loop */
 180         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 181         {
 182
 183             /* Get j neighbor index, and coordinate index */
 184             jnrA             = jjnr[jidx];
 185             jnrB             = jjnr[jidx+1];
 186             jnrC             = jjnr[jidx+2];
 187             jnrD             = jjnr[jidx+3];
 188             j_coord_offsetA  = DIM*jnrA;
 189             j_coord_offsetB  = DIM*jnrB;
 190             j_coord_offsetC  = DIM*jnrC;
 191             j_coord_offsetD  = DIM*jnrD;
 192
 193             /* load j atom coordinates */
 194             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 195                                               x+j_coord_offsetC,x+j_coord_offsetD,
 196                                               &jx0,&jy0,&jz0);
 197
 198             /* Calculate displacement vector */
 199             dx00             = _mm_sub_ps(ix0,jx0);
 200             dy00             = _mm_sub_ps(iy0,jy0);
 201             dz00             = _mm_sub_ps(iz0,jz0);
 202             dx10             = _mm_sub_ps(ix1,jx0);
 203             dy10             = _mm_sub_ps(iy1,jy0);
 204             dz10             = _mm_sub_ps(iz1,jz0);
 205             dx20             = _mm_sub_ps(ix2,jx0);
 206             dy20             = _mm_sub_ps(iy2,jy0);
 207             dz20             = _mm_sub_ps(iz2,jz0);
 208
 209             /* Calculate squared distance and things based on it */
 210             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 211             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 212             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 213
 214             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 215             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 216             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 217
 218             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 219             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 220             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 221
 222             /* Load parameters for j particles */
 223             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 224                                                               charge+jnrC+0,charge+jnrD+0);
 225             vdwjidx0A        = 2*vdwtype[jnrA+0];
 226             vdwjidx0B        = 2*vdwtype[jnrB+0];
 227             vdwjidx0C        = 2*vdwtype[jnrC+0];
 228             vdwjidx0D        = 2*vdwtype[jnrD+0];
 229
 230             fjx0             = _mm_setzero_ps();
 231             fjy0             = _mm_setzero_ps();
 232             fjz0             = _mm_setzero_ps();
 233
 234             /**************************
 235              * CALCULATE INTERACTIONS *
 236              **************************/
 237
 238             /* Compute parameters for interactions between i and j atoms */
 239             qq00             = _mm_mul_ps(iq0,jq0);
 240             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 241                                          vdwparam+vdwioffset0+vdwjidx0B,
 242                                          vdwparam+vdwioffset0+vdwjidx0C,
 243                                          vdwparam+vdwioffset0+vdwjidx0D,
 244                                          &c6_00,&c12_00);
 245
 246             /* COULOMB ELECTROSTATICS */
 247             velec            = _mm_mul_ps(qq00,rinv00);
 248             felec            = _mm_mul_ps(velec,rinvsq00);
 249
 250             /* LENNARD-JONES DISPERSION/REPULSION */
 251
 252             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 253             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 254             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 255             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 256             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 257
 258             /* Update potential sum for this i atom from the interaction with this j atom. */
 259             velecsum         = _mm_add_ps(velecsum,velec);
 260             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 261
 262             fscal            = _mm_add_ps(felec,fvdw);
 263
 264             /* Calculate temporary vectorial force */
 265             tx               = _mm_mul_ps(fscal,dx00);
 266             ty               = _mm_mul_ps(fscal,dy00);
 267             tz               = _mm_mul_ps(fscal,dz00);
 268
 269             /* Update vectorial force */
 270             fix0             = _mm_add_ps(fix0,tx);
 271             fiy0             = _mm_add_ps(fiy0,ty);
 272             fiz0             = _mm_add_ps(fiz0,tz);
 273
 274             fjx0             = _mm_add_ps(fjx0,tx);
 275             fjy0             = _mm_add_ps(fjy0,ty);
 276             fjz0             = _mm_add_ps(fjz0,tz);
 277
 278             /**************************
 279              * CALCULATE INTERACTIONS *
 280              **************************/
 281
 282             /* Compute parameters for interactions between i and j atoms */
 283             qq10             = _mm_mul_ps(iq1,jq0);
 284
 285             /* COULOMB ELECTROSTATICS */
 286             velec            = _mm_mul_ps(qq10,rinv10);
 287             felec            = _mm_mul_ps(velec,rinvsq10);
 288
 289             /* Update potential sum for this i atom from the interaction with this j atom. */
 290             velecsum         = _mm_add_ps(velecsum,velec);
 291
 292             fscal            = felec;
 293
 294             /* Calculate temporary vectorial force */
 295             tx               = _mm_mul_ps(fscal,dx10);
 296             ty               = _mm_mul_ps(fscal,dy10);
 297             tz               = _mm_mul_ps(fscal,dz10);
 298
 299             /* Update vectorial force */
 300             fix1             = _mm_add_ps(fix1,tx);
 301             fiy1             = _mm_add_ps(fiy1,ty);
 302             fiz1             = _mm_add_ps(fiz1,tz);
 303
 304             fjx0             = _mm_add_ps(fjx0,tx);
 305             fjy0             = _mm_add_ps(fjy0,ty);
 306             fjz0             = _mm_add_ps(fjz0,tz);
 307
 308             /**************************
 309              * CALCULATE INTERACTIONS *
 310              **************************/
 311
 312             /* Compute parameters for interactions between i and j atoms */
 313             qq20             = _mm_mul_ps(iq2,jq0);
 314
 315             /* COULOMB ELECTROSTATICS */
 316             velec            = _mm_mul_ps(qq20,rinv20);
 317             felec            = _mm_mul_ps(velec,rinvsq20);
 318
 319             /* Update potential sum for this i atom from the interaction with this j atom. */
 320             velecsum         = _mm_add_ps(velecsum,velec);
 321
 322             fscal            = felec;
 323
 324             /* Calculate temporary vectorial force */
 325             tx               = _mm_mul_ps(fscal,dx20);
 326             ty               = _mm_mul_ps(fscal,dy20);
 327             tz               = _mm_mul_ps(fscal,dz20);
 328
 329             /* Update vectorial force */
 330             fix2             = _mm_add_ps(fix2,tx);
 331             fiy2             = _mm_add_ps(fiy2,ty);
 332             fiz2             = _mm_add_ps(fiz2,tz);
 333
 334             fjx0             = _mm_add_ps(fjx0,tx);
 335             fjy0             = _mm_add_ps(fjy0,ty);
 336             fjz0             = _mm_add_ps(fjz0,tz);
 337
 338             fjptrA             = f+j_coord_offsetA;
 339             fjptrB             = f+j_coord_offsetB;
 340             fjptrC             = f+j_coord_offsetC;
 341             fjptrD             = f+j_coord_offsetD;
 342
 343             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 344
 345             /* Inner loop uses 96 flops */
 346         }
 347
 348         if(jidx<j_index_end)
 349         {
 350
 351             /* Get j neighbor index, and coordinate index */
 352             jnrlistA         = jjnr[jidx];
 353             jnrlistB         = jjnr[jidx+1];
 354             jnrlistC         = jjnr[jidx+2];
 355             jnrlistD         = jjnr[jidx+3];
 356             /* Sign of each element will be negative for non-real atoms.
 357              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 358              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 359              */
 360             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 361             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 362             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 363             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 364             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 365             j_coord_offsetA  = DIM*jnrA;
 366             j_coord_offsetB  = DIM*jnrB;
 367             j_coord_offsetC  = DIM*jnrC;
 368             j_coord_offsetD  = DIM*jnrD;
 369
 370             /* load j atom coordinates */
 371             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 372                                               x+j_coord_offsetC,x+j_coord_offsetD,
 373                                               &jx0,&jy0,&jz0);
 374
 375             /* Calculate displacement vector */
 376             dx00             = _mm_sub_ps(ix0,jx0);
 377             dy00             = _mm_sub_ps(iy0,jy0);
 378             dz00             = _mm_sub_ps(iz0,jz0);
 379             dx10             = _mm_sub_ps(ix1,jx0);
 380             dy10             = _mm_sub_ps(iy1,jy0);
 381             dz10             = _mm_sub_ps(iz1,jz0);
 382             dx20             = _mm_sub_ps(ix2,jx0);
 383             dy20             = _mm_sub_ps(iy2,jy0);
 384             dz20             = _mm_sub_ps(iz2,jz0);
 385
 386             /* Calculate squared distance and things based on it */
 387             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 388             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 389             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 390
 391             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 392             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 393             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 394
 395             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 396             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 397             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 398
 399             /* Load parameters for j particles */
 400             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 401                                                               charge+jnrC+0,charge+jnrD+0);
 402             vdwjidx0A        = 2*vdwtype[jnrA+0];
 403             vdwjidx0B        = 2*vdwtype[jnrB+0];
 404             vdwjidx0C        = 2*vdwtype[jnrC+0];
 405             vdwjidx0D        = 2*vdwtype[jnrD+0];
 406
 407             fjx0             = _mm_setzero_ps();
 408             fjy0             = _mm_setzero_ps();
 409             fjz0             = _mm_setzero_ps();
 410
 411             /**************************
 412              * CALCULATE INTERACTIONS *
 413              **************************/
 414
 415             /* Compute parameters for interactions between i and j atoms */
 416             qq00             = _mm_mul_ps(iq0,jq0);
 417             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 418                                          vdwparam+vdwioffset0+vdwjidx0B,
 419                                          vdwparam+vdwioffset0+vdwjidx0C,
 420                                          vdwparam+vdwioffset0+vdwjidx0D,
 421                                          &c6_00,&c12_00);
 422
 423             /* COULOMB ELECTROSTATICS */
 424             velec            = _mm_mul_ps(qq00,rinv00);
 425             felec            = _mm_mul_ps(velec,rinvsq00);
 426
 427             /* LENNARD-JONES DISPERSION/REPULSION */
 428
 429             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 430             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 431             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 432             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 433             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 434
 435             /* Update potential sum for this i atom from the interaction with this j atom. */
 436             velec            = _mm_andnot_ps(dummy_mask,velec);
 437             velecsum         = _mm_add_ps(velecsum,velec);
 438             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 439             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 440
 441             fscal            = _mm_add_ps(felec,fvdw);
 442
 443             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 444
 445             /* Calculate temporary vectorial force */
 446             tx               = _mm_mul_ps(fscal,dx00);
 447             ty               = _mm_mul_ps(fscal,dy00);
 448             tz               = _mm_mul_ps(fscal,dz00);
 449
 450             /* Update vectorial force */
 451             fix0             = _mm_add_ps(fix0,tx);
 452             fiy0             = _mm_add_ps(fiy0,ty);
 453             fiz0             = _mm_add_ps(fiz0,tz);
 454
 455             fjx0             = _mm_add_ps(fjx0,tx);
 456             fjy0             = _mm_add_ps(fjy0,ty);
 457             fjz0             = _mm_add_ps(fjz0,tz);
 458
 459             /**************************
 460              * CALCULATE INTERACTIONS *
 461              **************************/
 462
 463             /* Compute parameters for interactions between i and j atoms */
 464             qq10             = _mm_mul_ps(iq1,jq0);
 465
 466             /* COULOMB ELECTROSTATICS */
 467             velec            = _mm_mul_ps(qq10,rinv10);
 468             felec            = _mm_mul_ps(velec,rinvsq10);
 469
 470             /* Update potential sum for this i atom from the interaction with this j atom. */
 471             velec            = _mm_andnot_ps(dummy_mask,velec);
 472             velecsum         = _mm_add_ps(velecsum,velec);
 473
 474             fscal            = felec;
 475
 476             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 477
 478             /* Calculate temporary vectorial force */
 479             tx               = _mm_mul_ps(fscal,dx10);
 480             ty               = _mm_mul_ps(fscal,dy10);
 481             tz               = _mm_mul_ps(fscal,dz10);
 482
 483             /* Update vectorial force */
 484             fix1             = _mm_add_ps(fix1,tx);
 485             fiy1             = _mm_add_ps(fiy1,ty);
 486             fiz1             = _mm_add_ps(fiz1,tz);
 487
 488             fjx0             = _mm_add_ps(fjx0,tx);
 489             fjy0             = _mm_add_ps(fjy0,ty);
 490             fjz0             = _mm_add_ps(fjz0,tz);
 491
 492             /**************************
 493              * CALCULATE INTERACTIONS *
 494              **************************/
 495
 496             /* Compute parameters for interactions between i and j atoms */
 497             qq20             = _mm_mul_ps(iq2,jq0);
 498
 499             /* COULOMB ELECTROSTATICS */
 500             velec            = _mm_mul_ps(qq20,rinv20);
 501             felec            = _mm_mul_ps(velec,rinvsq20);
 502
 503             /* Update potential sum for this i atom from the interaction with this j atom. */
 504             velec            = _mm_andnot_ps(dummy_mask,velec);
 505             velecsum         = _mm_add_ps(velecsum,velec);
 506
 507             fscal            = felec;
 508
 509             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 510
 511             /* Calculate temporary vectorial force */
 512             tx               = _mm_mul_ps(fscal,dx20);
 513             ty               = _mm_mul_ps(fscal,dy20);
 514             tz               = _mm_mul_ps(fscal,dz20);
 515
 516             /* Update vectorial force */
 517             fix2             = _mm_add_ps(fix2,tx);
 518             fiy2             = _mm_add_ps(fiy2,ty);
 519             fiz2             = _mm_add_ps(fiz2,tz);
 520
 521             fjx0             = _mm_add_ps(fjx0,tx);
 522             fjy0             = _mm_add_ps(fjy0,ty);
 523             fjz0             = _mm_add_ps(fjz0,tz);
 524
 525             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 526             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 527             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 528             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 529
 530             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 531
 532             /* Inner loop uses 96 flops */
 533         }
 534
 535         /* End of innermost loop */
 536
 537         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 538                                               f+i_coord_offset,fshift+i_shift_offset);
 539
 540         ggid                        = gid[iidx];
 541         /* Update potential energies */
 542         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 543         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 544
 545         /* Increment number of inner iterations */
 546         inneriter                  += j_index_end - j_index_start;
 547
 548         /* Outer loop uses 20 flops */
 549     }
 550
 551     /* Increment number of outer iterations */
 552     outeriter        += nri;
 553
 554     /* Update outer/inner flops */
 555
 556     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*96);
 557 }
 558 /*
 559  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_sse4_1_single
 560  * Electrostatics interaction: Coulomb
 561  * VdW interaction:            LennardJones
 562  * Geometry:                   Water3-Particle
 563  * Calculate force/pot:        Force
 564  */
 565 void
 566 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_sse4_1_single
 567                     (t_nblist                    * gmx_restrict       nlist,
 568                      rvec                        * gmx_restrict          xx,
 569                      rvec                        * gmx_restrict          ff,
 570                      t_forcerec                  * gmx_restrict          fr,
 571                      t_mdatoms                   * gmx_restrict     mdatoms,
 572                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 573                      t_nrnb                      * gmx_restrict        nrnb)
 574 {
 575     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 576      * just 0 for non-waters.
 577      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 578      * jnr indices corresponding to data put in the four positions in the SIMD register.
 579      */
 580     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 581     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 582     int              jnrA,jnrB,jnrC,jnrD;
 583     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 584     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 585     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 586     real             rcutoff_scalar;
 587     real             *shiftvec,*fshift,*x,*f;
 588     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 589     real             scratch[4*DIM];
 590     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 591     int              vdwioffset0;
 592     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 593     int              vdwioffset1;
 594     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 595     int              vdwioffset2;
 596     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 597     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 598     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 599     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 600     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 601     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 602     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 603     real             *charge;
 604     int              nvdwtype;
 605     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 606     int              *vdwtype;
 607     real             *vdwparam;
 608     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 609     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 610     __m128           dummy_mask,cutoff_mask;
 611     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 612     __m128           one     = _mm_set1_ps(1.0);
 613     __m128           two     = _mm_set1_ps(2.0);
 614     x                = xx[0];
 615     f                = ff[0];
 616
 617     nri              = nlist->nri;
 618     iinr             = nlist->iinr;
 619     jindex           = nlist->jindex;
 620     jjnr             = nlist->jjnr;
 621     shiftidx         = nlist->shift;
 622     gid              = nlist->gid;
 623     shiftvec         = fr->shift_vec[0];
 624     fshift           = fr->fshift[0];
 625     facel            = _mm_set1_ps(fr->epsfac);
 626     charge           = mdatoms->chargeA;
 627     nvdwtype         = fr->ntype;
 628     vdwparam         = fr->nbfp;
 629     vdwtype          = mdatoms->typeA;
 630
 631     /* Setup water-specific parameters */
 632     inr              = nlist->iinr[0];
 633     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 634     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 635     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 636     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 637
 638     /* Avoid stupid compiler warnings */
 639     jnrA = jnrB = jnrC = jnrD = 0;
 640     j_coord_offsetA = 0;
 641     j_coord_offsetB = 0;
 642     j_coord_offsetC = 0;
 643     j_coord_offsetD = 0;
 644
 645     outeriter        = 0;
 646     inneriter        = 0;
 647
 648     for(iidx=0;iidx<4*DIM;iidx++)
 649     {
 650         scratch[iidx] = 0.0;
 651     }
 652
 653     /* Start outer loop over neighborlists */
 654     for(iidx=0; iidx<nri; iidx++)
 655     {
 656         /* Load shift vector for this list */
 657         i_shift_offset   = DIM*shiftidx[iidx];
 658
 659         /* Load limits for loop over neighbors */
 660         j_index_start    = jindex[iidx];
 661         j_index_end      = jindex[iidx+1];
 662
 663         /* Get outer coordinate index */
 664         inr              = iinr[iidx];
 665         i_coord_offset   = DIM*inr;
 666
 667         /* Load i particle coords and add shift vector */
 668         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 669                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 670
 671         fix0             = _mm_setzero_ps();
 672         fiy0             = _mm_setzero_ps();
 673         fiz0             = _mm_setzero_ps();
 674         fix1             = _mm_setzero_ps();
 675         fiy1             = _mm_setzero_ps();
 676         fiz1             = _mm_setzero_ps();
 677         fix2             = _mm_setzero_ps();
 678         fiy2             = _mm_setzero_ps();
 679         fiz2             = _mm_setzero_ps();
 680
 681         /* Start inner kernel loop */
 682         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 683         {
 684
 685             /* Get j neighbor index, and coordinate index */
 686             jnrA             = jjnr[jidx];
 687             jnrB             = jjnr[jidx+1];
 688             jnrC             = jjnr[jidx+2];
 689             jnrD             = jjnr[jidx+3];
 690             j_coord_offsetA  = DIM*jnrA;
 691             j_coord_offsetB  = DIM*jnrB;
 692             j_coord_offsetC  = DIM*jnrC;
 693             j_coord_offsetD  = DIM*jnrD;
 694
 695             /* load j atom coordinates */
 696             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 697                                               x+j_coord_offsetC,x+j_coord_offsetD,
 698                                               &jx0,&jy0,&jz0);
 699
 700             /* Calculate displacement vector */
 701             dx00             = _mm_sub_ps(ix0,jx0);
 702             dy00             = _mm_sub_ps(iy0,jy0);
 703             dz00             = _mm_sub_ps(iz0,jz0);
 704             dx10             = _mm_sub_ps(ix1,jx0);
 705             dy10             = _mm_sub_ps(iy1,jy0);
 706             dz10             = _mm_sub_ps(iz1,jz0);
 707             dx20             = _mm_sub_ps(ix2,jx0);
 708             dy20             = _mm_sub_ps(iy2,jy0);
 709             dz20             = _mm_sub_ps(iz2,jz0);
 710
 711             /* Calculate squared distance and things based on it */
 712             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 713             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 714             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 715
 716             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 717             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 718             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 719
 720             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 721             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 722             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 723
 724             /* Load parameters for j particles */
 725             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 726                                                               charge+jnrC+0,charge+jnrD+0);
 727             vdwjidx0A        = 2*vdwtype[jnrA+0];
 728             vdwjidx0B        = 2*vdwtype[jnrB+0];
 729             vdwjidx0C        = 2*vdwtype[jnrC+0];
 730             vdwjidx0D        = 2*vdwtype[jnrD+0];
 731
 732             fjx0             = _mm_setzero_ps();
 733             fjy0             = _mm_setzero_ps();
 734             fjz0             = _mm_setzero_ps();
 735
 736             /**************************
 737              * CALCULATE INTERACTIONS *
 738              **************************/
 739
 740             /* Compute parameters for interactions between i and j atoms */
 741             qq00             = _mm_mul_ps(iq0,jq0);
 742             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 743                                          vdwparam+vdwioffset0+vdwjidx0B,
 744                                          vdwparam+vdwioffset0+vdwjidx0C,
 745                                          vdwparam+vdwioffset0+vdwjidx0D,
 746                                          &c6_00,&c12_00);
 747
 748             /* COULOMB ELECTROSTATICS */
 749             velec            = _mm_mul_ps(qq00,rinv00);
 750             felec            = _mm_mul_ps(velec,rinvsq00);
 751
 752             /* LENNARD-JONES DISPERSION/REPULSION */
 753
 754             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 755             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 756
 757             fscal            = _mm_add_ps(felec,fvdw);
 758
 759             /* Calculate temporary vectorial force */
 760             tx               = _mm_mul_ps(fscal,dx00);
 761             ty               = _mm_mul_ps(fscal,dy00);
 762             tz               = _mm_mul_ps(fscal,dz00);
 763
 764             /* Update vectorial force */
 765             fix0             = _mm_add_ps(fix0,tx);
 766             fiy0             = _mm_add_ps(fiy0,ty);
 767             fiz0             = _mm_add_ps(fiz0,tz);
 768
 769             fjx0             = _mm_add_ps(fjx0,tx);
 770             fjy0             = _mm_add_ps(fjy0,ty);
 771             fjz0             = _mm_add_ps(fjz0,tz);
 772
 773             /**************************
 774              * CALCULATE INTERACTIONS *
 775              **************************/
 776
 777             /* Compute parameters for interactions between i and j atoms */
 778             qq10             = _mm_mul_ps(iq1,jq0);
 779
 780             /* COULOMB ELECTROSTATICS */
 781             velec            = _mm_mul_ps(qq10,rinv10);
 782             felec            = _mm_mul_ps(velec,rinvsq10);
 783
 784             fscal            = felec;
 785
 786             /* Calculate temporary vectorial force */
 787             tx               = _mm_mul_ps(fscal,dx10);
 788             ty               = _mm_mul_ps(fscal,dy10);
 789             tz               = _mm_mul_ps(fscal,dz10);
 790
 791             /* Update vectorial force */
 792             fix1             = _mm_add_ps(fix1,tx);
 793             fiy1             = _mm_add_ps(fiy1,ty);
 794             fiz1             = _mm_add_ps(fiz1,tz);
 795
 796             fjx0             = _mm_add_ps(fjx0,tx);
 797             fjy0             = _mm_add_ps(fjy0,ty);
 798             fjz0             = _mm_add_ps(fjz0,tz);
 799
 800             /**************************
 801              * CALCULATE INTERACTIONS *
 802              **************************/
 803
 804             /* Compute parameters for interactions between i and j atoms */
 805             qq20             = _mm_mul_ps(iq2,jq0);
 806
 807             /* COULOMB ELECTROSTATICS */
 808             velec            = _mm_mul_ps(qq20,rinv20);
 809             felec            = _mm_mul_ps(velec,rinvsq20);
 810
 811             fscal            = felec;
 812
 813             /* Calculate temporary vectorial force */
 814             tx               = _mm_mul_ps(fscal,dx20);
 815             ty               = _mm_mul_ps(fscal,dy20);
 816             tz               = _mm_mul_ps(fscal,dz20);
 817
 818             /* Update vectorial force */
 819             fix2             = _mm_add_ps(fix2,tx);
 820             fiy2             = _mm_add_ps(fiy2,ty);
 821             fiz2             = _mm_add_ps(fiz2,tz);
 822
 823             fjx0             = _mm_add_ps(fjx0,tx);
 824             fjy0             = _mm_add_ps(fjy0,ty);
 825             fjz0             = _mm_add_ps(fjz0,tz);
 826
 827             fjptrA             = f+j_coord_offsetA;
 828             fjptrB             = f+j_coord_offsetB;
 829             fjptrC             = f+j_coord_offsetC;
 830             fjptrD             = f+j_coord_offsetD;
 831
 832             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 833
 834             /* Inner loop uses 88 flops */
 835         }
 836
 837         if(jidx<j_index_end)
 838         {
 839
 840             /* Get j neighbor index, and coordinate index */
 841             jnrlistA         = jjnr[jidx];
 842             jnrlistB         = jjnr[jidx+1];
 843             jnrlistC         = jjnr[jidx+2];
 844             jnrlistD         = jjnr[jidx+3];
 845             /* Sign of each element will be negative for non-real atoms.
 846              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 847              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 848              */
 849             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 850             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 851             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 852             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 853             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 854             j_coord_offsetA  = DIM*jnrA;
 855             j_coord_offsetB  = DIM*jnrB;
 856             j_coord_offsetC  = DIM*jnrC;
 857             j_coord_offsetD  = DIM*jnrD;
 858
 859             /* load j atom coordinates */
 860             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 861                                               x+j_coord_offsetC,x+j_coord_offsetD,
 862                                               &jx0,&jy0,&jz0);
 863
 864             /* Calculate displacement vector */
 865             dx00             = _mm_sub_ps(ix0,jx0);
 866             dy00             = _mm_sub_ps(iy0,jy0);
 867             dz00             = _mm_sub_ps(iz0,jz0);
 868             dx10             = _mm_sub_ps(ix1,jx0);
 869             dy10             = _mm_sub_ps(iy1,jy0);
 870             dz10             = _mm_sub_ps(iz1,jz0);
 871             dx20             = _mm_sub_ps(ix2,jx0);
 872             dy20             = _mm_sub_ps(iy2,jy0);
 873             dz20             = _mm_sub_ps(iz2,jz0);
 874
 875             /* Calculate squared distance and things based on it */
 876             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 877             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 878             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 879
 880             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 881             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 882             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 883
 884             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 885             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 886             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 887
 888             /* Load parameters for j particles */
 889             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 890                                                               charge+jnrC+0,charge+jnrD+0);
 891             vdwjidx0A        = 2*vdwtype[jnrA+0];
 892             vdwjidx0B        = 2*vdwtype[jnrB+0];
 893             vdwjidx0C        = 2*vdwtype[jnrC+0];
 894             vdwjidx0D        = 2*vdwtype[jnrD+0];
 895
 896             fjx0             = _mm_setzero_ps();
 897             fjy0             = _mm_setzero_ps();
 898             fjz0             = _mm_setzero_ps();
 899
 900             /**************************
 901              * CALCULATE INTERACTIONS *
 902              **************************/
 903
 904             /* Compute parameters for interactions between i and j atoms */
 905             qq00             = _mm_mul_ps(iq0,jq0);
 906             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 907                                          vdwparam+vdwioffset0+vdwjidx0B,
 908                                          vdwparam+vdwioffset0+vdwjidx0C,
 909                                          vdwparam+vdwioffset0+vdwjidx0D,
 910                                          &c6_00,&c12_00);
 911
 912             /* COULOMB ELECTROSTATICS */
 913             velec            = _mm_mul_ps(qq00,rinv00);
 914             felec            = _mm_mul_ps(velec,rinvsq00);
 915
 916             /* LENNARD-JONES DISPERSION/REPULSION */
 917
 918             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 919             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 920
 921             fscal            = _mm_add_ps(felec,fvdw);
 922
 923             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 924
 925             /* Calculate temporary vectorial force */
 926             tx               = _mm_mul_ps(fscal,dx00);
 927             ty               = _mm_mul_ps(fscal,dy00);
 928             tz               = _mm_mul_ps(fscal,dz00);
 929
 930             /* Update vectorial force */
 931             fix0             = _mm_add_ps(fix0,tx);
 932             fiy0             = _mm_add_ps(fiy0,ty);
 933             fiz0             = _mm_add_ps(fiz0,tz);
 934
 935             fjx0             = _mm_add_ps(fjx0,tx);
 936             fjy0             = _mm_add_ps(fjy0,ty);
 937             fjz0             = _mm_add_ps(fjz0,tz);
 938
 939             /**************************
 940              * CALCULATE INTERACTIONS *
 941              **************************/
 942
 943             /* Compute parameters for interactions between i and j atoms */
 944             qq10             = _mm_mul_ps(iq1,jq0);
 945
 946             /* COULOMB ELECTROSTATICS */
 947             velec            = _mm_mul_ps(qq10,rinv10);
 948             felec            = _mm_mul_ps(velec,rinvsq10);
 949
 950             fscal            = felec;
 951
 952             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 953
 954             /* Calculate temporary vectorial force */
 955             tx               = _mm_mul_ps(fscal,dx10);
 956             ty               = _mm_mul_ps(fscal,dy10);
 957             tz               = _mm_mul_ps(fscal,dz10);
 958
 959             /* Update vectorial force */
 960             fix1             = _mm_add_ps(fix1,tx);
 961             fiy1             = _mm_add_ps(fiy1,ty);
 962             fiz1             = _mm_add_ps(fiz1,tz);
 963
 964             fjx0             = _mm_add_ps(fjx0,tx);
 965             fjy0             = _mm_add_ps(fjy0,ty);
 966             fjz0             = _mm_add_ps(fjz0,tz);
 967
 968             /**************************
 969              * CALCULATE INTERACTIONS *
 970              **************************/
 971
 972             /* Compute parameters for interactions between i and j atoms */
 973             qq20             = _mm_mul_ps(iq2,jq0);
 974
 975             /* COULOMB ELECTROSTATICS */
 976             velec            = _mm_mul_ps(qq20,rinv20);
 977             felec            = _mm_mul_ps(velec,rinvsq20);
 978
 979             fscal            = felec;
 980
 981             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 982
 983             /* Calculate temporary vectorial force */
 984             tx               = _mm_mul_ps(fscal,dx20);
 985             ty               = _mm_mul_ps(fscal,dy20);
 986             tz               = _mm_mul_ps(fscal,dz20);
 987
 988             /* Update vectorial force */
 989             fix2             = _mm_add_ps(fix2,tx);
 990             fiy2             = _mm_add_ps(fiy2,ty);
 991             fiz2             = _mm_add_ps(fiz2,tz);
 992
 993             fjx0             = _mm_add_ps(fjx0,tx);
 994             fjy0             = _mm_add_ps(fjy0,ty);
 995             fjz0             = _mm_add_ps(fjz0,tz);
 996
 997             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 998             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 999             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1000             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1001
1002             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1003
1004             /* Inner loop uses 88 flops */
1005         }
1006
1007         /* End of innermost loop */
1008
1009         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1010                                               f+i_coord_offset,fshift+i_shift_offset);
1011
1012         /* Increment number of inner iterations */
1013         inneriter                  += j_index_end - j_index_start;
1014
1015         /* Outer loop uses 18 flops */
1016     }
1017
1018     /* Increment number of outer iterations */
1019     outeriter        += nri;
1020
1021     /* Update outer/inner flops */
1022
1023     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*88);
1024 }