src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecCoul_VdwLJ_GeomW3P1_sse2_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_sse2_single.h"
  34 #include "kernelutil_x86_sse2_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW3P1_VF_sse2_single
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_VF_sse2_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             shX,shY,shZ,rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwioffset1;
  69     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  70     int              vdwioffset2;
  71     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  72     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  73     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  75     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  76     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  77     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     int              nvdwtype;
  80     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  81     int              *vdwtype;
  82     real             *vdwparam;
  83     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  84     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  85     __m128           dummy_mask,cutoff_mask;
  86     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  87     __m128           one     = _mm_set1_ps(1.0);
  88     __m128           two     = _mm_set1_ps(2.0);
  89     x                = xx[0];
  90     f                = ff[0];
  91
  92     nri              = nlist->nri;
  93     iinr             = nlist->iinr;
  94     jindex           = nlist->jindex;
  95     jjnr             = nlist->jjnr;
  96     shiftidx         = nlist->shift;
  97     gid              = nlist->gid;
  98     shiftvec         = fr->shift_vec[0];
  99     fshift           = fr->fshift[0];
 100     facel            = _mm_set1_ps(fr->epsfac);
 101     charge           = mdatoms->chargeA;
 102     nvdwtype         = fr->ntype;
 103     vdwparam         = fr->nbfp;
 104     vdwtype          = mdatoms->typeA;
 105
 106     /* Setup water-specific parameters */
 107     inr              = nlist->iinr[0];
 108     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 109     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 110     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 111     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 112
 113     /* Avoid stupid compiler warnings */
 114     jnrA = jnrB = jnrC = jnrD = 0;
 115     j_coord_offsetA = 0;
 116     j_coord_offsetB = 0;
 117     j_coord_offsetC = 0;
 118     j_coord_offsetD = 0;
 119
 120     outeriter        = 0;
 121     inneriter        = 0;
 122
 123     /* Start outer loop over neighborlists */
 124     for(iidx=0; iidx<nri; iidx++)
 125     {
 126         /* Load shift vector for this list */
 127         i_shift_offset   = DIM*shiftidx[iidx];
 128         shX              = shiftvec[i_shift_offset+XX];
 129         shY              = shiftvec[i_shift_offset+YY];
 130         shZ              = shiftvec[i_shift_offset+ZZ];
 131
 132         /* Load limits for loop over neighbors */
 133         j_index_start    = jindex[iidx];
 134         j_index_end      = jindex[iidx+1];
 135
 136         /* Get outer coordinate index */
 137         inr              = iinr[iidx];
 138         i_coord_offset   = DIM*inr;
 139
 140         /* Load i particle coords and add shift vector */
 141         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 142         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 143         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 144         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 145         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 146         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 147         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 148         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 149         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 150
 151         fix0             = _mm_setzero_ps();
 152         fiy0             = _mm_setzero_ps();
 153         fiz0             = _mm_setzero_ps();
 154         fix1             = _mm_setzero_ps();
 155         fiy1             = _mm_setzero_ps();
 156         fiz1             = _mm_setzero_ps();
 157         fix2             = _mm_setzero_ps();
 158         fiy2             = _mm_setzero_ps();
 159         fiz2             = _mm_setzero_ps();
 160
 161         /* Reset potential sums */
 162         velecsum         = _mm_setzero_ps();
 163         vvdwsum          = _mm_setzero_ps();
 164
 165         /* Start inner kernel loop */
 166         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 167         {
 168
 169             /* Get j neighbor index, and coordinate index */
 170             jnrA             = jjnr[jidx];
 171             jnrB             = jjnr[jidx+1];
 172             jnrC             = jjnr[jidx+2];
 173             jnrD             = jjnr[jidx+3];
 174
 175             j_coord_offsetA  = DIM*jnrA;
 176             j_coord_offsetB  = DIM*jnrB;
 177             j_coord_offsetC  = DIM*jnrC;
 178             j_coord_offsetD  = DIM*jnrD;
 179
 180             /* load j atom coordinates */
 181             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 182                                               x+j_coord_offsetC,x+j_coord_offsetD,
 183                                               &jx0,&jy0,&jz0);
 184
 185             /* Calculate displacement vector */
 186             dx00             = _mm_sub_ps(ix0,jx0);
 187             dy00             = _mm_sub_ps(iy0,jy0);
 188             dz00             = _mm_sub_ps(iz0,jz0);
 189             dx10             = _mm_sub_ps(ix1,jx0);
 190             dy10             = _mm_sub_ps(iy1,jy0);
 191             dz10             = _mm_sub_ps(iz1,jz0);
 192             dx20             = _mm_sub_ps(ix2,jx0);
 193             dy20             = _mm_sub_ps(iy2,jy0);
 194             dz20             = _mm_sub_ps(iz2,jz0);
 195
 196             /* Calculate squared distance and things based on it */
 197             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 198             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 199             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 200
 201             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 202             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 203             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 204
 205             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 206             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 207             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 208
 209             /* Load parameters for j particles */
 210             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 211                                                               charge+jnrC+0,charge+jnrD+0);
 212             vdwjidx0A        = 2*vdwtype[jnrA+0];
 213             vdwjidx0B        = 2*vdwtype[jnrB+0];
 214             vdwjidx0C        = 2*vdwtype[jnrC+0];
 215             vdwjidx0D        = 2*vdwtype[jnrD+0];
 216
 217             /**************************
 218              * CALCULATE INTERACTIONS *
 219              **************************/
 220
 221             /* Compute parameters for interactions between i and j atoms */
 222             qq00             = _mm_mul_ps(iq0,jq0);
 223             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 224                                          vdwparam+vdwioffset0+vdwjidx0B,
 225                                          vdwparam+vdwioffset0+vdwjidx0C,
 226                                          vdwparam+vdwioffset0+vdwjidx0D,
 227                                          &c6_00,&c12_00);
 228
 229             /* COULOMB ELECTROSTATICS */
 230             velec            = _mm_mul_ps(qq00,rinv00);
 231             felec            = _mm_mul_ps(velec,rinvsq00);
 232
 233             /* LENNARD-JONES DISPERSION/REPULSION */
 234
 235             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 236             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 237             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 238             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 239             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 240
 241             /* Update potential sum for this i atom from the interaction with this j atom. */
 242             velecsum         = _mm_add_ps(velecsum,velec);
 243             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 244
 245             fscal            = _mm_add_ps(felec,fvdw);
 246
 247             /* Calculate temporary vectorial force */
 248             tx               = _mm_mul_ps(fscal,dx00);
 249             ty               = _mm_mul_ps(fscal,dy00);
 250             tz               = _mm_mul_ps(fscal,dz00);
 251
 252             /* Update vectorial force */
 253             fix0             = _mm_add_ps(fix0,tx);
 254             fiy0             = _mm_add_ps(fiy0,ty);
 255             fiz0             = _mm_add_ps(fiz0,tz);
 256
 257             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 258                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 259                                                    tx,ty,tz);
 260
 261             /**************************
 262              * CALCULATE INTERACTIONS *
 263              **************************/
 264
 265             /* Compute parameters for interactions between i and j atoms */
 266             qq10             = _mm_mul_ps(iq1,jq0);
 267
 268             /* COULOMB ELECTROSTATICS */
 269             velec            = _mm_mul_ps(qq10,rinv10);
 270             felec            = _mm_mul_ps(velec,rinvsq10);
 271
 272             /* Update potential sum for this i atom from the interaction with this j atom. */
 273             velecsum         = _mm_add_ps(velecsum,velec);
 274
 275             fscal            = felec;
 276
 277             /* Calculate temporary vectorial force */
 278             tx               = _mm_mul_ps(fscal,dx10);
 279             ty               = _mm_mul_ps(fscal,dy10);
 280             tz               = _mm_mul_ps(fscal,dz10);
 281
 282             /* Update vectorial force */
 283             fix1             = _mm_add_ps(fix1,tx);
 284             fiy1             = _mm_add_ps(fiy1,ty);
 285             fiz1             = _mm_add_ps(fiz1,tz);
 286
 287             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 288                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 289                                                    tx,ty,tz);
 290
 291             /**************************
 292              * CALCULATE INTERACTIONS *
 293              **************************/
 294
 295             /* Compute parameters for interactions between i and j atoms */
 296             qq20             = _mm_mul_ps(iq2,jq0);
 297
 298             /* COULOMB ELECTROSTATICS */
 299             velec            = _mm_mul_ps(qq20,rinv20);
 300             felec            = _mm_mul_ps(velec,rinvsq20);
 301
 302             /* Update potential sum for this i atom from the interaction with this j atom. */
 303             velecsum         = _mm_add_ps(velecsum,velec);
 304
 305             fscal            = felec;
 306
 307             /* Calculate temporary vectorial force */
 308             tx               = _mm_mul_ps(fscal,dx20);
 309             ty               = _mm_mul_ps(fscal,dy20);
 310             tz               = _mm_mul_ps(fscal,dz20);
 311
 312             /* Update vectorial force */
 313             fix2             = _mm_add_ps(fix2,tx);
 314             fiy2             = _mm_add_ps(fiy2,ty);
 315             fiz2             = _mm_add_ps(fiz2,tz);
 316
 317             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 318                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 319                                                    tx,ty,tz);
 320
 321             /* Inner loop uses 96 flops */
 322         }
 323
 324         if(jidx<j_index_end)
 325         {
 326
 327             /* Get j neighbor index, and coordinate index */
 328             jnrA             = jjnr[jidx];
 329             jnrB             = jjnr[jidx+1];
 330             jnrC             = jjnr[jidx+2];
 331             jnrD             = jjnr[jidx+3];
 332
 333             /* Sign of each element will be negative for non-real atoms.
 334              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 335              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 336              */
 337             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 338             jnrA       = (jnrA>=0) ? jnrA : 0;
 339             jnrB       = (jnrB>=0) ? jnrB : 0;
 340             jnrC       = (jnrC>=0) ? jnrC : 0;
 341             jnrD       = (jnrD>=0) ? jnrD : 0;
 342
 343             j_coord_offsetA  = DIM*jnrA;
 344             j_coord_offsetB  = DIM*jnrB;
 345             j_coord_offsetC  = DIM*jnrC;
 346             j_coord_offsetD  = DIM*jnrD;
 347
 348             /* load j atom coordinates */
 349             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 350                                               x+j_coord_offsetC,x+j_coord_offsetD,
 351                                               &jx0,&jy0,&jz0);
 352
 353             /* Calculate displacement vector */
 354             dx00             = _mm_sub_ps(ix0,jx0);
 355             dy00             = _mm_sub_ps(iy0,jy0);
 356             dz00             = _mm_sub_ps(iz0,jz0);
 357             dx10             = _mm_sub_ps(ix1,jx0);
 358             dy10             = _mm_sub_ps(iy1,jy0);
 359             dz10             = _mm_sub_ps(iz1,jz0);
 360             dx20             = _mm_sub_ps(ix2,jx0);
 361             dy20             = _mm_sub_ps(iy2,jy0);
 362             dz20             = _mm_sub_ps(iz2,jz0);
 363
 364             /* Calculate squared distance and things based on it */
 365             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 366             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 367             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 368
 369             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 370             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 371             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 372
 373             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 374             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 375             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 376
 377             /* Load parameters for j particles */
 378             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 379                                                               charge+jnrC+0,charge+jnrD+0);
 380             vdwjidx0A        = 2*vdwtype[jnrA+0];
 381             vdwjidx0B        = 2*vdwtype[jnrB+0];
 382             vdwjidx0C        = 2*vdwtype[jnrC+0];
 383             vdwjidx0D        = 2*vdwtype[jnrD+0];
 384
 385             /**************************
 386              * CALCULATE INTERACTIONS *
 387              **************************/
 388
 389             /* Compute parameters for interactions between i and j atoms */
 390             qq00             = _mm_mul_ps(iq0,jq0);
 391             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 392                                          vdwparam+vdwioffset0+vdwjidx0B,
 393                                          vdwparam+vdwioffset0+vdwjidx0C,
 394                                          vdwparam+vdwioffset0+vdwjidx0D,
 395                                          &c6_00,&c12_00);
 396
 397             /* COULOMB ELECTROSTATICS */
 398             velec            = _mm_mul_ps(qq00,rinv00);
 399             felec            = _mm_mul_ps(velec,rinvsq00);
 400
 401             /* LENNARD-JONES DISPERSION/REPULSION */
 402
 403             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 404             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 405             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 406             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 407             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 408
 409             /* Update potential sum for this i atom from the interaction with this j atom. */
 410             velec            = _mm_andnot_ps(dummy_mask,velec);
 411             velecsum         = _mm_add_ps(velecsum,velec);
 412             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 413             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 414
 415             fscal            = _mm_add_ps(felec,fvdw);
 416
 417             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 418
 419             /* Calculate temporary vectorial force */
 420             tx               = _mm_mul_ps(fscal,dx00);
 421             ty               = _mm_mul_ps(fscal,dy00);
 422             tz               = _mm_mul_ps(fscal,dz00);
 423
 424             /* Update vectorial force */
 425             fix0             = _mm_add_ps(fix0,tx);
 426             fiy0             = _mm_add_ps(fiy0,ty);
 427             fiz0             = _mm_add_ps(fiz0,tz);
 428
 429             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 430                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 431                                                    tx,ty,tz);
 432
 433             /**************************
 434              * CALCULATE INTERACTIONS *
 435              **************************/
 436
 437             /* Compute parameters for interactions between i and j atoms */
 438             qq10             = _mm_mul_ps(iq1,jq0);
 439
 440             /* COULOMB ELECTROSTATICS */
 441             velec            = _mm_mul_ps(qq10,rinv10);
 442             felec            = _mm_mul_ps(velec,rinvsq10);
 443
 444             /* Update potential sum for this i atom from the interaction with this j atom. */
 445             velec            = _mm_andnot_ps(dummy_mask,velec);
 446             velecsum         = _mm_add_ps(velecsum,velec);
 447
 448             fscal            = felec;
 449
 450             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 451
 452             /* Calculate temporary vectorial force */
 453             tx               = _mm_mul_ps(fscal,dx10);
 454             ty               = _mm_mul_ps(fscal,dy10);
 455             tz               = _mm_mul_ps(fscal,dz10);
 456
 457             /* Update vectorial force */
 458             fix1             = _mm_add_ps(fix1,tx);
 459             fiy1             = _mm_add_ps(fiy1,ty);
 460             fiz1             = _mm_add_ps(fiz1,tz);
 461
 462             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 463                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 464                                                    tx,ty,tz);
 465
 466             /**************************
 467              * CALCULATE INTERACTIONS *
 468              **************************/
 469
 470             /* Compute parameters for interactions between i and j atoms */
 471             qq20             = _mm_mul_ps(iq2,jq0);
 472
 473             /* COULOMB ELECTROSTATICS */
 474             velec            = _mm_mul_ps(qq20,rinv20);
 475             felec            = _mm_mul_ps(velec,rinvsq20);
 476
 477             /* Update potential sum for this i atom from the interaction with this j atom. */
 478             velec            = _mm_andnot_ps(dummy_mask,velec);
 479             velecsum         = _mm_add_ps(velecsum,velec);
 480
 481             fscal            = felec;
 482
 483             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 484
 485             /* Calculate temporary vectorial force */
 486             tx               = _mm_mul_ps(fscal,dx20);
 487             ty               = _mm_mul_ps(fscal,dy20);
 488             tz               = _mm_mul_ps(fscal,dz20);
 489
 490             /* Update vectorial force */
 491             fix2             = _mm_add_ps(fix2,tx);
 492             fiy2             = _mm_add_ps(fiy2,ty);
 493             fiz2             = _mm_add_ps(fiz2,tz);
 494
 495             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 496                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 497                                                    tx,ty,tz);
 498
 499             /* Inner loop uses 96 flops */
 500         }
 501
 502         /* End of innermost loop */
 503
 504         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 505                                               f+i_coord_offset,fshift+i_shift_offset);
 506
 507         ggid                        = gid[iidx];
 508         /* Update potential energies */
 509         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 510         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 511
 512         /* Increment number of inner iterations */
 513         inneriter                  += j_index_end - j_index_start;
 514
 515         /* Outer loop uses 29 flops */
 516     }
 517
 518     /* Increment number of outer iterations */
 519     outeriter        += nri;
 520
 521     /* Update outer/inner flops */
 522
 523     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*29 + inneriter*96);
 524 }
 525 /*
 526  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_sse2_single
 527  * Electrostatics interaction: Coulomb
 528  * VdW interaction:            LennardJones
 529  * Geometry:                   Water3-Particle
 530  * Calculate force/pot:        Force
 531  */
 532 void
 533 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_sse2_single
 534                     (t_nblist * gmx_restrict                nlist,
 535                      rvec * gmx_restrict                    xx,
 536                      rvec * gmx_restrict                    ff,
 537                      t_forcerec * gmx_restrict              fr,
 538                      t_mdatoms * gmx_restrict               mdatoms,
 539                      nb_kernel_data_t * gmx_restrict        kernel_data,
 540                      t_nrnb * gmx_restrict                  nrnb)
 541 {
 542     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 543      * just 0 for non-waters.
 544      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 545      * jnr indices corresponding to data put in the four positions in the SIMD register.
 546      */
 547     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 548     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 549     int              jnrA,jnrB,jnrC,jnrD;
 550     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 551     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 552     real             shX,shY,shZ,rcutoff_scalar;
 553     real             *shiftvec,*fshift,*x,*f;
 554     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 555     int              vdwioffset0;
 556     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 557     int              vdwioffset1;
 558     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 559     int              vdwioffset2;
 560     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 561     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 562     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 563     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 564     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 565     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 566     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 567     real             *charge;
 568     int              nvdwtype;
 569     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 570     int              *vdwtype;
 571     real             *vdwparam;
 572     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 573     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 574     __m128           dummy_mask,cutoff_mask;
 575     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 576     __m128           one     = _mm_set1_ps(1.0);
 577     __m128           two     = _mm_set1_ps(2.0);
 578     x                = xx[0];
 579     f                = ff[0];
 580
 581     nri              = nlist->nri;
 582     iinr             = nlist->iinr;
 583     jindex           = nlist->jindex;
 584     jjnr             = nlist->jjnr;
 585     shiftidx         = nlist->shift;
 586     gid              = nlist->gid;
 587     shiftvec         = fr->shift_vec[0];
 588     fshift           = fr->fshift[0];
 589     facel            = _mm_set1_ps(fr->epsfac);
 590     charge           = mdatoms->chargeA;
 591     nvdwtype         = fr->ntype;
 592     vdwparam         = fr->nbfp;
 593     vdwtype          = mdatoms->typeA;
 594
 595     /* Setup water-specific parameters */
 596     inr              = nlist->iinr[0];
 597     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 598     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 599     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 600     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 601
 602     /* Avoid stupid compiler warnings */
 603     jnrA = jnrB = jnrC = jnrD = 0;
 604     j_coord_offsetA = 0;
 605     j_coord_offsetB = 0;
 606     j_coord_offsetC = 0;
 607     j_coord_offsetD = 0;
 608
 609     outeriter        = 0;
 610     inneriter        = 0;
 611
 612     /* Start outer loop over neighborlists */
 613     for(iidx=0; iidx<nri; iidx++)
 614     {
 615         /* Load shift vector for this list */
 616         i_shift_offset   = DIM*shiftidx[iidx];
 617         shX              = shiftvec[i_shift_offset+XX];
 618         shY              = shiftvec[i_shift_offset+YY];
 619         shZ              = shiftvec[i_shift_offset+ZZ];
 620
 621         /* Load limits for loop over neighbors */
 622         j_index_start    = jindex[iidx];
 623         j_index_end      = jindex[iidx+1];
 624
 625         /* Get outer coordinate index */
 626         inr              = iinr[iidx];
 627         i_coord_offset   = DIM*inr;
 628
 629         /* Load i particle coords and add shift vector */
 630         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 631         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 632         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 633         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 634         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 635         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 636         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 637         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 638         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 639
 640         fix0             = _mm_setzero_ps();
 641         fiy0             = _mm_setzero_ps();
 642         fiz0             = _mm_setzero_ps();
 643         fix1             = _mm_setzero_ps();
 644         fiy1             = _mm_setzero_ps();
 645         fiz1             = _mm_setzero_ps();
 646         fix2             = _mm_setzero_ps();
 647         fiy2             = _mm_setzero_ps();
 648         fiz2             = _mm_setzero_ps();
 649
 650         /* Start inner kernel loop */
 651         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 652         {
 653
 654             /* Get j neighbor index, and coordinate index */
 655             jnrA             = jjnr[jidx];
 656             jnrB             = jjnr[jidx+1];
 657             jnrC             = jjnr[jidx+2];
 658             jnrD             = jjnr[jidx+3];
 659
 660             j_coord_offsetA  = DIM*jnrA;
 661             j_coord_offsetB  = DIM*jnrB;
 662             j_coord_offsetC  = DIM*jnrC;
 663             j_coord_offsetD  = DIM*jnrD;
 664
 665             /* load j atom coordinates */
 666             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 667                                               x+j_coord_offsetC,x+j_coord_offsetD,
 668                                               &jx0,&jy0,&jz0);
 669
 670             /* Calculate displacement vector */
 671             dx00             = _mm_sub_ps(ix0,jx0);
 672             dy00             = _mm_sub_ps(iy0,jy0);
 673             dz00             = _mm_sub_ps(iz0,jz0);
 674             dx10             = _mm_sub_ps(ix1,jx0);
 675             dy10             = _mm_sub_ps(iy1,jy0);
 676             dz10             = _mm_sub_ps(iz1,jz0);
 677             dx20             = _mm_sub_ps(ix2,jx0);
 678             dy20             = _mm_sub_ps(iy2,jy0);
 679             dz20             = _mm_sub_ps(iz2,jz0);
 680
 681             /* Calculate squared distance and things based on it */
 682             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 683             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 684             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 685
 686             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 687             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 688             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 689
 690             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 691             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 692             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 693
 694             /* Load parameters for j particles */
 695             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 696                                                               charge+jnrC+0,charge+jnrD+0);
 697             vdwjidx0A        = 2*vdwtype[jnrA+0];
 698             vdwjidx0B        = 2*vdwtype[jnrB+0];
 699             vdwjidx0C        = 2*vdwtype[jnrC+0];
 700             vdwjidx0D        = 2*vdwtype[jnrD+0];
 701
 702             /**************************
 703              * CALCULATE INTERACTIONS *
 704              **************************/
 705
 706             /* Compute parameters for interactions between i and j atoms */
 707             qq00             = _mm_mul_ps(iq0,jq0);
 708             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 709                                          vdwparam+vdwioffset0+vdwjidx0B,
 710                                          vdwparam+vdwioffset0+vdwjidx0C,
 711                                          vdwparam+vdwioffset0+vdwjidx0D,
 712                                          &c6_00,&c12_00);
 713
 714             /* COULOMB ELECTROSTATICS */
 715             velec            = _mm_mul_ps(qq00,rinv00);
 716             felec            = _mm_mul_ps(velec,rinvsq00);
 717
 718             /* LENNARD-JONES DISPERSION/REPULSION */
 719
 720             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 721             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 722
 723             fscal            = _mm_add_ps(felec,fvdw);
 724
 725             /* Calculate temporary vectorial force */
 726             tx               = _mm_mul_ps(fscal,dx00);
 727             ty               = _mm_mul_ps(fscal,dy00);
 728             tz               = _mm_mul_ps(fscal,dz00);
 729
 730             /* Update vectorial force */
 731             fix0             = _mm_add_ps(fix0,tx);
 732             fiy0             = _mm_add_ps(fiy0,ty);
 733             fiz0             = _mm_add_ps(fiz0,tz);
 734
 735             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 736                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 737                                                    tx,ty,tz);
 738
 739             /**************************
 740              * CALCULATE INTERACTIONS *
 741              **************************/
 742
 743             /* Compute parameters for interactions between i and j atoms */
 744             qq10             = _mm_mul_ps(iq1,jq0);
 745
 746             /* COULOMB ELECTROSTATICS */
 747             velec            = _mm_mul_ps(qq10,rinv10);
 748             felec            = _mm_mul_ps(velec,rinvsq10);
 749
 750             fscal            = felec;
 751
 752             /* Calculate temporary vectorial force */
 753             tx               = _mm_mul_ps(fscal,dx10);
 754             ty               = _mm_mul_ps(fscal,dy10);
 755             tz               = _mm_mul_ps(fscal,dz10);
 756
 757             /* Update vectorial force */
 758             fix1             = _mm_add_ps(fix1,tx);
 759             fiy1             = _mm_add_ps(fiy1,ty);
 760             fiz1             = _mm_add_ps(fiz1,tz);
 761
 762             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 763                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 764                                                    tx,ty,tz);
 765
 766             /**************************
 767              * CALCULATE INTERACTIONS *
 768              **************************/
 769
 770             /* Compute parameters for interactions between i and j atoms */
 771             qq20             = _mm_mul_ps(iq2,jq0);
 772
 773             /* COULOMB ELECTROSTATICS */
 774             velec            = _mm_mul_ps(qq20,rinv20);
 775             felec            = _mm_mul_ps(velec,rinvsq20);
 776
 777             fscal            = felec;
 778
 779             /* Calculate temporary vectorial force */
 780             tx               = _mm_mul_ps(fscal,dx20);
 781             ty               = _mm_mul_ps(fscal,dy20);
 782             tz               = _mm_mul_ps(fscal,dz20);
 783
 784             /* Update vectorial force */
 785             fix2             = _mm_add_ps(fix2,tx);
 786             fiy2             = _mm_add_ps(fiy2,ty);
 787             fiz2             = _mm_add_ps(fiz2,tz);
 788
 789             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 790                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 791                                                    tx,ty,tz);
 792
 793             /* Inner loop uses 88 flops */
 794         }
 795
 796         if(jidx<j_index_end)
 797         {
 798
 799             /* Get j neighbor index, and coordinate index */
 800             jnrA             = jjnr[jidx];
 801             jnrB             = jjnr[jidx+1];
 802             jnrC             = jjnr[jidx+2];
 803             jnrD             = jjnr[jidx+3];
 804
 805             /* Sign of each element will be negative for non-real atoms.
 806              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 807              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 808              */
 809             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 810             jnrA       = (jnrA>=0) ? jnrA : 0;
 811             jnrB       = (jnrB>=0) ? jnrB : 0;
 812             jnrC       = (jnrC>=0) ? jnrC : 0;
 813             jnrD       = (jnrD>=0) ? jnrD : 0;
 814
 815             j_coord_offsetA  = DIM*jnrA;
 816             j_coord_offsetB  = DIM*jnrB;
 817             j_coord_offsetC  = DIM*jnrC;
 818             j_coord_offsetD  = DIM*jnrD;
 819
 820             /* load j atom coordinates */
 821             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 822                                               x+j_coord_offsetC,x+j_coord_offsetD,
 823                                               &jx0,&jy0,&jz0);
 824
 825             /* Calculate displacement vector */
 826             dx00             = _mm_sub_ps(ix0,jx0);
 827             dy00             = _mm_sub_ps(iy0,jy0);
 828             dz00             = _mm_sub_ps(iz0,jz0);
 829             dx10             = _mm_sub_ps(ix1,jx0);
 830             dy10             = _mm_sub_ps(iy1,jy0);
 831             dz10             = _mm_sub_ps(iz1,jz0);
 832             dx20             = _mm_sub_ps(ix2,jx0);
 833             dy20             = _mm_sub_ps(iy2,jy0);
 834             dz20             = _mm_sub_ps(iz2,jz0);
 835
 836             /* Calculate squared distance and things based on it */
 837             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 838             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 839             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 840
 841             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 842             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 843             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 844
 845             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 846             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 847             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 848
 849             /* Load parameters for j particles */
 850             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 851                                                               charge+jnrC+0,charge+jnrD+0);
 852             vdwjidx0A        = 2*vdwtype[jnrA+0];
 853             vdwjidx0B        = 2*vdwtype[jnrB+0];
 854             vdwjidx0C        = 2*vdwtype[jnrC+0];
 855             vdwjidx0D        = 2*vdwtype[jnrD+0];
 856
 857             /**************************
 858              * CALCULATE INTERACTIONS *
 859              **************************/
 860
 861             /* Compute parameters for interactions between i and j atoms */
 862             qq00             = _mm_mul_ps(iq0,jq0);
 863             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 864                                          vdwparam+vdwioffset0+vdwjidx0B,
 865                                          vdwparam+vdwioffset0+vdwjidx0C,
 866                                          vdwparam+vdwioffset0+vdwjidx0D,
 867                                          &c6_00,&c12_00);
 868
 869             /* COULOMB ELECTROSTATICS */
 870             velec            = _mm_mul_ps(qq00,rinv00);
 871             felec            = _mm_mul_ps(velec,rinvsq00);
 872
 873             /* LENNARD-JONES DISPERSION/REPULSION */
 874
 875             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 876             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 877
 878             fscal            = _mm_add_ps(felec,fvdw);
 879
 880             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 881
 882             /* Calculate temporary vectorial force */
 883             tx               = _mm_mul_ps(fscal,dx00);
 884             ty               = _mm_mul_ps(fscal,dy00);
 885             tz               = _mm_mul_ps(fscal,dz00);
 886
 887             /* Update vectorial force */
 888             fix0             = _mm_add_ps(fix0,tx);
 889             fiy0             = _mm_add_ps(fiy0,ty);
 890             fiz0             = _mm_add_ps(fiz0,tz);
 891
 892             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 893                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 894                                                    tx,ty,tz);
 895
 896             /**************************
 897              * CALCULATE INTERACTIONS *
 898              **************************/
 899
 900             /* Compute parameters for interactions between i and j atoms */
 901             qq10             = _mm_mul_ps(iq1,jq0);
 902
 903             /* COULOMB ELECTROSTATICS */
 904             velec            = _mm_mul_ps(qq10,rinv10);
 905             felec            = _mm_mul_ps(velec,rinvsq10);
 906
 907             fscal            = felec;
 908
 909             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 910
 911             /* Calculate temporary vectorial force */
 912             tx               = _mm_mul_ps(fscal,dx10);
 913             ty               = _mm_mul_ps(fscal,dy10);
 914             tz               = _mm_mul_ps(fscal,dz10);
 915
 916             /* Update vectorial force */
 917             fix1             = _mm_add_ps(fix1,tx);
 918             fiy1             = _mm_add_ps(fiy1,ty);
 919             fiz1             = _mm_add_ps(fiz1,tz);
 920
 921             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 922                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 923                                                    tx,ty,tz);
 924
 925             /**************************
 926              * CALCULATE INTERACTIONS *
 927              **************************/
 928
 929             /* Compute parameters for interactions between i and j atoms */
 930             qq20             = _mm_mul_ps(iq2,jq0);
 931
 932             /* COULOMB ELECTROSTATICS */
 933             velec            = _mm_mul_ps(qq20,rinv20);
 934             felec            = _mm_mul_ps(velec,rinvsq20);
 935
 936             fscal            = felec;
 937
 938             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 939
 940             /* Calculate temporary vectorial force */
 941             tx               = _mm_mul_ps(fscal,dx20);
 942             ty               = _mm_mul_ps(fscal,dy20);
 943             tz               = _mm_mul_ps(fscal,dz20);
 944
 945             /* Update vectorial force */
 946             fix2             = _mm_add_ps(fix2,tx);
 947             fiy2             = _mm_add_ps(fiy2,ty);
 948             fiz2             = _mm_add_ps(fiz2,tz);
 949
 950             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 951                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 952                                                    tx,ty,tz);
 953
 954             /* Inner loop uses 88 flops */
 955         }
 956
 957         /* End of innermost loop */
 958
 959         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 960                                               f+i_coord_offset,fshift+i_shift_offset);
 961
 962         /* Increment number of inner iterations */
 963         inneriter                  += j_index_end - j_index_start;
 964
 965         /* Outer loop uses 27 flops */
 966     }
 967
 968     /* Increment number of outer iterations */
 969     outeriter        += nri;
 970
 971     /* Update outer/inner flops */
 972
 973     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*27 + inneriter*88);
 974 }