src/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              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwioffset1;
  72     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  73     int              vdwioffset2;
  74     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  75     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  76     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  77     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  78     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  79     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  80     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     int              nvdwtype;
  83     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  84     int              *vdwtype;
  85     real             *vdwparam;
  86     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  87     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  88     __m128           dummy_mask,cutoff_mask;
  89     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  90     __m128           one     = _mm_set1_ps(1.0);
  91     __m128           two     = _mm_set1_ps(2.0);
  92     x                = xx[0];
  93     f                = ff[0];
  94
  95     nri              = nlist->nri;
  96     iinr             = nlist->iinr;
  97     jindex           = nlist->jindex;
  98     jjnr             = nlist->jjnr;
  99     shiftidx         = nlist->shift;
 100     gid              = nlist->gid;
 101     shiftvec         = fr->shift_vec[0];
 102     fshift           = fr->fshift[0];
 103     facel            = _mm_set1_ps(fr->epsfac);
 104     charge           = mdatoms->chargeA;
 105     nvdwtype         = fr->ntype;
 106     vdwparam         = fr->nbfp;
 107     vdwtype          = mdatoms->typeA;
 108
 109     /* Setup water-specific parameters */
 110     inr              = nlist->iinr[0];
 111     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 112     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 113     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 114     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 115
 116     /* Avoid stupid compiler warnings */
 117     jnrA = jnrB = jnrC = jnrD = 0;
 118     j_coord_offsetA = 0;
 119     j_coord_offsetB = 0;
 120     j_coord_offsetC = 0;
 121     j_coord_offsetD = 0;
 122
 123     outeriter        = 0;
 124     inneriter        = 0;
 125
 126     for(iidx=0;iidx<4*DIM;iidx++)
 127     {
 128         scratch[iidx] = 0.0;
 129     }
 130
 131     /* Start outer loop over neighborlists */
 132     for(iidx=0; iidx<nri; iidx++)
 133     {
 134         /* Load shift vector for this list */
 135         i_shift_offset   = DIM*shiftidx[iidx];
 136
 137         /* Load limits for loop over neighbors */
 138         j_index_start    = jindex[iidx];
 139         j_index_end      = jindex[iidx+1];
 140
 141         /* Get outer coordinate index */
 142         inr              = iinr[iidx];
 143         i_coord_offset   = DIM*inr;
 144
 145         /* Load i particle coords and add shift vector */
 146         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 147                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 148
 149         fix0             = _mm_setzero_ps();
 150         fiy0             = _mm_setzero_ps();
 151         fiz0             = _mm_setzero_ps();
 152         fix1             = _mm_setzero_ps();
 153         fiy1             = _mm_setzero_ps();
 154         fiz1             = _mm_setzero_ps();
 155         fix2             = _mm_setzero_ps();
 156         fiy2             = _mm_setzero_ps();
 157         fiz2             = _mm_setzero_ps();
 158
 159         /* Reset potential sums */
 160         velecsum         = _mm_setzero_ps();
 161         vvdwsum          = _mm_setzero_ps();
 162
 163         /* Start inner kernel loop */
 164         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 165         {
 166
 167             /* Get j neighbor index, and coordinate index */
 168             jnrA             = jjnr[jidx];
 169             jnrB             = jjnr[jidx+1];
 170             jnrC             = jjnr[jidx+2];
 171             jnrD             = jjnr[jidx+3];
 172             j_coord_offsetA  = DIM*jnrA;
 173             j_coord_offsetB  = DIM*jnrB;
 174             j_coord_offsetC  = DIM*jnrC;
 175             j_coord_offsetD  = DIM*jnrD;
 176
 177             /* load j atom coordinates */
 178             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 179                                               x+j_coord_offsetC,x+j_coord_offsetD,
 180                                               &jx0,&jy0,&jz0);
 181
 182             /* Calculate displacement vector */
 183             dx00             = _mm_sub_ps(ix0,jx0);
 184             dy00             = _mm_sub_ps(iy0,jy0);
 185             dz00             = _mm_sub_ps(iz0,jz0);
 186             dx10             = _mm_sub_ps(ix1,jx0);
 187             dy10             = _mm_sub_ps(iy1,jy0);
 188             dz10             = _mm_sub_ps(iz1,jz0);
 189             dx20             = _mm_sub_ps(ix2,jx0);
 190             dy20             = _mm_sub_ps(iy2,jy0);
 191             dz20             = _mm_sub_ps(iz2,jz0);
 192
 193             /* Calculate squared distance and things based on it */
 194             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 195             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 196             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 197
 198             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 199             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 200             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 201
 202             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 203             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 204             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 205
 206             /* Load parameters for j particles */
 207             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 208                                                               charge+jnrC+0,charge+jnrD+0);
 209             vdwjidx0A        = 2*vdwtype[jnrA+0];
 210             vdwjidx0B        = 2*vdwtype[jnrB+0];
 211             vdwjidx0C        = 2*vdwtype[jnrC+0];
 212             vdwjidx0D        = 2*vdwtype[jnrD+0];
 213
 214             fjx0             = _mm_setzero_ps();
 215             fjy0             = _mm_setzero_ps();
 216             fjz0             = _mm_setzero_ps();
 217
 218             /**************************
 219              * CALCULATE INTERACTIONS *
 220              **************************/
 221
 222             /* Compute parameters for interactions between i and j atoms */
 223             qq00             = _mm_mul_ps(iq0,jq0);
 224             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 225                                          vdwparam+vdwioffset0+vdwjidx0B,
 226                                          vdwparam+vdwioffset0+vdwjidx0C,
 227                                          vdwparam+vdwioffset0+vdwjidx0D,
 228                                          &c6_00,&c12_00);
 229
 230             /* COULOMB ELECTROSTATICS */
 231             velec            = _mm_mul_ps(qq00,rinv00);
 232             felec            = _mm_mul_ps(velec,rinvsq00);
 233
 234             /* LENNARD-JONES DISPERSION/REPULSION */
 235
 236             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 237             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 238             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 239             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 240             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 241
 242             /* Update potential sum for this i atom from the interaction with this j atom. */
 243             velecsum         = _mm_add_ps(velecsum,velec);
 244             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 245
 246             fscal            = _mm_add_ps(felec,fvdw);
 247
 248             /* Calculate temporary vectorial force */
 249             tx               = _mm_mul_ps(fscal,dx00);
 250             ty               = _mm_mul_ps(fscal,dy00);
 251             tz               = _mm_mul_ps(fscal,dz00);
 252
 253             /* Update vectorial force */
 254             fix0             = _mm_add_ps(fix0,tx);
 255             fiy0             = _mm_add_ps(fiy0,ty);
 256             fiz0             = _mm_add_ps(fiz0,tz);
 257
 258             fjx0             = _mm_add_ps(fjx0,tx);
 259             fjy0             = _mm_add_ps(fjy0,ty);
 260             fjz0             = _mm_add_ps(fjz0,tz);
 261
 262             /**************************
 263              * CALCULATE INTERACTIONS *
 264              **************************/
 265
 266             /* Compute parameters for interactions between i and j atoms */
 267             qq10             = _mm_mul_ps(iq1,jq0);
 268
 269             /* COULOMB ELECTROSTATICS */
 270             velec            = _mm_mul_ps(qq10,rinv10);
 271             felec            = _mm_mul_ps(velec,rinvsq10);
 272
 273             /* Update potential sum for this i atom from the interaction with this j atom. */
 274             velecsum         = _mm_add_ps(velecsum,velec);
 275
 276             fscal            = felec;
 277
 278             /* Calculate temporary vectorial force */
 279             tx               = _mm_mul_ps(fscal,dx10);
 280             ty               = _mm_mul_ps(fscal,dy10);
 281             tz               = _mm_mul_ps(fscal,dz10);
 282
 283             /* Update vectorial force */
 284             fix1             = _mm_add_ps(fix1,tx);
 285             fiy1             = _mm_add_ps(fiy1,ty);
 286             fiz1             = _mm_add_ps(fiz1,tz);
 287
 288             fjx0             = _mm_add_ps(fjx0,tx);
 289             fjy0             = _mm_add_ps(fjy0,ty);
 290             fjz0             = _mm_add_ps(fjz0,tz);
 291
 292             /**************************
 293              * CALCULATE INTERACTIONS *
 294              **************************/
 295
 296             /* Compute parameters for interactions between i and j atoms */
 297             qq20             = _mm_mul_ps(iq2,jq0);
 298
 299             /* COULOMB ELECTROSTATICS */
 300             velec            = _mm_mul_ps(qq20,rinv20);
 301             felec            = _mm_mul_ps(velec,rinvsq20);
 302
 303             /* Update potential sum for this i atom from the interaction with this j atom. */
 304             velecsum         = _mm_add_ps(velecsum,velec);
 305
 306             fscal            = felec;
 307
 308             /* Calculate temporary vectorial force */
 309             tx               = _mm_mul_ps(fscal,dx20);
 310             ty               = _mm_mul_ps(fscal,dy20);
 311             tz               = _mm_mul_ps(fscal,dz20);
 312
 313             /* Update vectorial force */
 314             fix2             = _mm_add_ps(fix2,tx);
 315             fiy2             = _mm_add_ps(fiy2,ty);
 316             fiz2             = _mm_add_ps(fiz2,tz);
 317
 318             fjx0             = _mm_add_ps(fjx0,tx);
 319             fjy0             = _mm_add_ps(fjy0,ty);
 320             fjz0             = _mm_add_ps(fjz0,tz);
 321
 322             fjptrA             = f+j_coord_offsetA;
 323             fjptrB             = f+j_coord_offsetB;
 324             fjptrC             = f+j_coord_offsetC;
 325             fjptrD             = f+j_coord_offsetD;
 326
 327             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 328
 329             /* Inner loop uses 96 flops */
 330         }
 331
 332         if(jidx<j_index_end)
 333         {
 334
 335             /* Get j neighbor index, and coordinate index */
 336             jnrlistA         = jjnr[jidx];
 337             jnrlistB         = jjnr[jidx+1];
 338             jnrlistC         = jjnr[jidx+2];
 339             jnrlistD         = jjnr[jidx+3];
 340             /* Sign of each element will be negative for non-real atoms.
 341              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 342              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 343              */
 344             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 345             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 346             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 347             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 348             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 349             j_coord_offsetA  = DIM*jnrA;
 350             j_coord_offsetB  = DIM*jnrB;
 351             j_coord_offsetC  = DIM*jnrC;
 352             j_coord_offsetD  = DIM*jnrD;
 353
 354             /* load j atom coordinates */
 355             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 356                                               x+j_coord_offsetC,x+j_coord_offsetD,
 357                                               &jx0,&jy0,&jz0);
 358
 359             /* Calculate displacement vector */
 360             dx00             = _mm_sub_ps(ix0,jx0);
 361             dy00             = _mm_sub_ps(iy0,jy0);
 362             dz00             = _mm_sub_ps(iz0,jz0);
 363             dx10             = _mm_sub_ps(ix1,jx0);
 364             dy10             = _mm_sub_ps(iy1,jy0);
 365             dz10             = _mm_sub_ps(iz1,jz0);
 366             dx20             = _mm_sub_ps(ix2,jx0);
 367             dy20             = _mm_sub_ps(iy2,jy0);
 368             dz20             = _mm_sub_ps(iz2,jz0);
 369
 370             /* Calculate squared distance and things based on it */
 371             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 372             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 373             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 374
 375             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 376             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 377             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 378
 379             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 380             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 381             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 382
 383             /* Load parameters for j particles */
 384             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 385                                                               charge+jnrC+0,charge+jnrD+0);
 386             vdwjidx0A        = 2*vdwtype[jnrA+0];
 387             vdwjidx0B        = 2*vdwtype[jnrB+0];
 388             vdwjidx0C        = 2*vdwtype[jnrC+0];
 389             vdwjidx0D        = 2*vdwtype[jnrD+0];
 390
 391             fjx0             = _mm_setzero_ps();
 392             fjy0             = _mm_setzero_ps();
 393             fjz0             = _mm_setzero_ps();
 394
 395             /**************************
 396              * CALCULATE INTERACTIONS *
 397              **************************/
 398
 399             /* Compute parameters for interactions between i and j atoms */
 400             qq00             = _mm_mul_ps(iq0,jq0);
 401             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 402                                          vdwparam+vdwioffset0+vdwjidx0B,
 403                                          vdwparam+vdwioffset0+vdwjidx0C,
 404                                          vdwparam+vdwioffset0+vdwjidx0D,
 405                                          &c6_00,&c12_00);
 406
 407             /* COULOMB ELECTROSTATICS */
 408             velec            = _mm_mul_ps(qq00,rinv00);
 409             felec            = _mm_mul_ps(velec,rinvsq00);
 410
 411             /* LENNARD-JONES DISPERSION/REPULSION */
 412
 413             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 414             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 415             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 416             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 417             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 418
 419             /* Update potential sum for this i atom from the interaction with this j atom. */
 420             velec            = _mm_andnot_ps(dummy_mask,velec);
 421             velecsum         = _mm_add_ps(velecsum,velec);
 422             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 423             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 424
 425             fscal            = _mm_add_ps(felec,fvdw);
 426
 427             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 428
 429             /* Calculate temporary vectorial force */
 430             tx               = _mm_mul_ps(fscal,dx00);
 431             ty               = _mm_mul_ps(fscal,dy00);
 432             tz               = _mm_mul_ps(fscal,dz00);
 433
 434             /* Update vectorial force */
 435             fix0             = _mm_add_ps(fix0,tx);
 436             fiy0             = _mm_add_ps(fiy0,ty);
 437             fiz0             = _mm_add_ps(fiz0,tz);
 438
 439             fjx0             = _mm_add_ps(fjx0,tx);
 440             fjy0             = _mm_add_ps(fjy0,ty);
 441             fjz0             = _mm_add_ps(fjz0,tz);
 442
 443             /**************************
 444              * CALCULATE INTERACTIONS *
 445              **************************/
 446
 447             /* Compute parameters for interactions between i and j atoms */
 448             qq10             = _mm_mul_ps(iq1,jq0);
 449
 450             /* COULOMB ELECTROSTATICS */
 451             velec            = _mm_mul_ps(qq10,rinv10);
 452             felec            = _mm_mul_ps(velec,rinvsq10);
 453
 454             /* Update potential sum for this i atom from the interaction with this j atom. */
 455             velec            = _mm_andnot_ps(dummy_mask,velec);
 456             velecsum         = _mm_add_ps(velecsum,velec);
 457
 458             fscal            = felec;
 459
 460             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 461
 462             /* Calculate temporary vectorial force */
 463             tx               = _mm_mul_ps(fscal,dx10);
 464             ty               = _mm_mul_ps(fscal,dy10);
 465             tz               = _mm_mul_ps(fscal,dz10);
 466
 467             /* Update vectorial force */
 468             fix1             = _mm_add_ps(fix1,tx);
 469             fiy1             = _mm_add_ps(fiy1,ty);
 470             fiz1             = _mm_add_ps(fiz1,tz);
 471
 472             fjx0             = _mm_add_ps(fjx0,tx);
 473             fjy0             = _mm_add_ps(fjy0,ty);
 474             fjz0             = _mm_add_ps(fjz0,tz);
 475
 476             /**************************
 477              * CALCULATE INTERACTIONS *
 478              **************************/
 479
 480             /* Compute parameters for interactions between i and j atoms */
 481             qq20             = _mm_mul_ps(iq2,jq0);
 482
 483             /* COULOMB ELECTROSTATICS */
 484             velec            = _mm_mul_ps(qq20,rinv20);
 485             felec            = _mm_mul_ps(velec,rinvsq20);
 486
 487             /* Update potential sum for this i atom from the interaction with this j atom. */
 488             velec            = _mm_andnot_ps(dummy_mask,velec);
 489             velecsum         = _mm_add_ps(velecsum,velec);
 490
 491             fscal            = felec;
 492
 493             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 494
 495             /* Calculate temporary vectorial force */
 496             tx               = _mm_mul_ps(fscal,dx20);
 497             ty               = _mm_mul_ps(fscal,dy20);
 498             tz               = _mm_mul_ps(fscal,dz20);
 499
 500             /* Update vectorial force */
 501             fix2             = _mm_add_ps(fix2,tx);
 502             fiy2             = _mm_add_ps(fiy2,ty);
 503             fiz2             = _mm_add_ps(fiz2,tz);
 504
 505             fjx0             = _mm_add_ps(fjx0,tx);
 506             fjy0             = _mm_add_ps(fjy0,ty);
 507             fjz0             = _mm_add_ps(fjz0,tz);
 508
 509             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 510             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 511             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 512             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 513
 514             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 515
 516             /* Inner loop uses 96 flops */
 517         }
 518
 519         /* End of innermost loop */
 520
 521         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 522                                               f+i_coord_offset,fshift+i_shift_offset);
 523
 524         ggid                        = gid[iidx];
 525         /* Update potential energies */
 526         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 527         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 528
 529         /* Increment number of inner iterations */
 530         inneriter                  += j_index_end - j_index_start;
 531
 532         /* Outer loop uses 20 flops */
 533     }
 534
 535     /* Increment number of outer iterations */
 536     outeriter        += nri;
 537
 538     /* Update outer/inner flops */
 539
 540     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*96);
 541 }
 542 /*
 543  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_sse2_single
 544  * Electrostatics interaction: Coulomb
 545  * VdW interaction:            LennardJones
 546  * Geometry:                   Water3-Particle
 547  * Calculate force/pot:        Force
 548  */
 549 void
 550 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_sse2_single
 551                     (t_nblist * gmx_restrict                nlist,
 552                      rvec * gmx_restrict                    xx,
 553                      rvec * gmx_restrict                    ff,
 554                      t_forcerec * gmx_restrict              fr,
 555                      t_mdatoms * gmx_restrict               mdatoms,
 556                      nb_kernel_data_t * gmx_restrict        kernel_data,
 557                      t_nrnb * gmx_restrict                  nrnb)
 558 {
 559     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 560      * just 0 for non-waters.
 561      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 562      * jnr indices corresponding to data put in the four positions in the SIMD register.
 563      */
 564     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 565     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 566     int              jnrA,jnrB,jnrC,jnrD;
 567     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 568     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 569     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 570     real             rcutoff_scalar;
 571     real             *shiftvec,*fshift,*x,*f;
 572     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 573     real             scratch[4*DIM];
 574     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 575     int              vdwioffset0;
 576     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 577     int              vdwioffset1;
 578     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 579     int              vdwioffset2;
 580     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 581     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 582     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 583     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 584     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 585     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 586     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 587     real             *charge;
 588     int              nvdwtype;
 589     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 590     int              *vdwtype;
 591     real             *vdwparam;
 592     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 593     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 594     __m128           dummy_mask,cutoff_mask;
 595     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 596     __m128           one     = _mm_set1_ps(1.0);
 597     __m128           two     = _mm_set1_ps(2.0);
 598     x                = xx[0];
 599     f                = ff[0];
 600
 601     nri              = nlist->nri;
 602     iinr             = nlist->iinr;
 603     jindex           = nlist->jindex;
 604     jjnr             = nlist->jjnr;
 605     shiftidx         = nlist->shift;
 606     gid              = nlist->gid;
 607     shiftvec         = fr->shift_vec[0];
 608     fshift           = fr->fshift[0];
 609     facel            = _mm_set1_ps(fr->epsfac);
 610     charge           = mdatoms->chargeA;
 611     nvdwtype         = fr->ntype;
 612     vdwparam         = fr->nbfp;
 613     vdwtype          = mdatoms->typeA;
 614
 615     /* Setup water-specific parameters */
 616     inr              = nlist->iinr[0];
 617     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 618     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 619     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 620     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 621
 622     /* Avoid stupid compiler warnings */
 623     jnrA = jnrB = jnrC = jnrD = 0;
 624     j_coord_offsetA = 0;
 625     j_coord_offsetB = 0;
 626     j_coord_offsetC = 0;
 627     j_coord_offsetD = 0;
 628
 629     outeriter        = 0;
 630     inneriter        = 0;
 631
 632     for(iidx=0;iidx<4*DIM;iidx++)
 633     {
 634         scratch[iidx] = 0.0;
 635     }
 636
 637     /* Start outer loop over neighborlists */
 638     for(iidx=0; iidx<nri; iidx++)
 639     {
 640         /* Load shift vector for this list */
 641         i_shift_offset   = DIM*shiftidx[iidx];
 642
 643         /* Load limits for loop over neighbors */
 644         j_index_start    = jindex[iidx];
 645         j_index_end      = jindex[iidx+1];
 646
 647         /* Get outer coordinate index */
 648         inr              = iinr[iidx];
 649         i_coord_offset   = DIM*inr;
 650
 651         /* Load i particle coords and add shift vector */
 652         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 653                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 654
 655         fix0             = _mm_setzero_ps();
 656         fiy0             = _mm_setzero_ps();
 657         fiz0             = _mm_setzero_ps();
 658         fix1             = _mm_setzero_ps();
 659         fiy1             = _mm_setzero_ps();
 660         fiz1             = _mm_setzero_ps();
 661         fix2             = _mm_setzero_ps();
 662         fiy2             = _mm_setzero_ps();
 663         fiz2             = _mm_setzero_ps();
 664
 665         /* Start inner kernel loop */
 666         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 667         {
 668
 669             /* Get j neighbor index, and coordinate index */
 670             jnrA             = jjnr[jidx];
 671             jnrB             = jjnr[jidx+1];
 672             jnrC             = jjnr[jidx+2];
 673             jnrD             = jjnr[jidx+3];
 674             j_coord_offsetA  = DIM*jnrA;
 675             j_coord_offsetB  = DIM*jnrB;
 676             j_coord_offsetC  = DIM*jnrC;
 677             j_coord_offsetD  = DIM*jnrD;
 678
 679             /* load j atom coordinates */
 680             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 681                                               x+j_coord_offsetC,x+j_coord_offsetD,
 682                                               &jx0,&jy0,&jz0);
 683
 684             /* Calculate displacement vector */
 685             dx00             = _mm_sub_ps(ix0,jx0);
 686             dy00             = _mm_sub_ps(iy0,jy0);
 687             dz00             = _mm_sub_ps(iz0,jz0);
 688             dx10             = _mm_sub_ps(ix1,jx0);
 689             dy10             = _mm_sub_ps(iy1,jy0);
 690             dz10             = _mm_sub_ps(iz1,jz0);
 691             dx20             = _mm_sub_ps(ix2,jx0);
 692             dy20             = _mm_sub_ps(iy2,jy0);
 693             dz20             = _mm_sub_ps(iz2,jz0);
 694
 695             /* Calculate squared distance and things based on it */
 696             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 697             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 698             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 699
 700             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 701             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 702             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 703
 704             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 705             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 706             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 707
 708             /* Load parameters for j particles */
 709             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 710                                                               charge+jnrC+0,charge+jnrD+0);
 711             vdwjidx0A        = 2*vdwtype[jnrA+0];
 712             vdwjidx0B        = 2*vdwtype[jnrB+0];
 713             vdwjidx0C        = 2*vdwtype[jnrC+0];
 714             vdwjidx0D        = 2*vdwtype[jnrD+0];
 715
 716             fjx0             = _mm_setzero_ps();
 717             fjy0             = _mm_setzero_ps();
 718             fjz0             = _mm_setzero_ps();
 719
 720             /**************************
 721              * CALCULATE INTERACTIONS *
 722              **************************/
 723
 724             /* Compute parameters for interactions between i and j atoms */
 725             qq00             = _mm_mul_ps(iq0,jq0);
 726             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 727                                          vdwparam+vdwioffset0+vdwjidx0B,
 728                                          vdwparam+vdwioffset0+vdwjidx0C,
 729                                          vdwparam+vdwioffset0+vdwjidx0D,
 730                                          &c6_00,&c12_00);
 731
 732             /* COULOMB ELECTROSTATICS */
 733             velec            = _mm_mul_ps(qq00,rinv00);
 734             felec            = _mm_mul_ps(velec,rinvsq00);
 735
 736             /* LENNARD-JONES DISPERSION/REPULSION */
 737
 738             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 739             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 740
 741             fscal            = _mm_add_ps(felec,fvdw);
 742
 743             /* Calculate temporary vectorial force */
 744             tx               = _mm_mul_ps(fscal,dx00);
 745             ty               = _mm_mul_ps(fscal,dy00);
 746             tz               = _mm_mul_ps(fscal,dz00);
 747
 748             /* Update vectorial force */
 749             fix0             = _mm_add_ps(fix0,tx);
 750             fiy0             = _mm_add_ps(fiy0,ty);
 751             fiz0             = _mm_add_ps(fiz0,tz);
 752
 753             fjx0             = _mm_add_ps(fjx0,tx);
 754             fjy0             = _mm_add_ps(fjy0,ty);
 755             fjz0             = _mm_add_ps(fjz0,tz);
 756
 757             /**************************
 758              * CALCULATE INTERACTIONS *
 759              **************************/
 760
 761             /* Compute parameters for interactions between i and j atoms */
 762             qq10             = _mm_mul_ps(iq1,jq0);
 763
 764             /* COULOMB ELECTROSTATICS */
 765             velec            = _mm_mul_ps(qq10,rinv10);
 766             felec            = _mm_mul_ps(velec,rinvsq10);
 767
 768             fscal            = felec;
 769
 770             /* Calculate temporary vectorial force */
 771             tx               = _mm_mul_ps(fscal,dx10);
 772             ty               = _mm_mul_ps(fscal,dy10);
 773             tz               = _mm_mul_ps(fscal,dz10);
 774
 775             /* Update vectorial force */
 776             fix1             = _mm_add_ps(fix1,tx);
 777             fiy1             = _mm_add_ps(fiy1,ty);
 778             fiz1             = _mm_add_ps(fiz1,tz);
 779
 780             fjx0             = _mm_add_ps(fjx0,tx);
 781             fjy0             = _mm_add_ps(fjy0,ty);
 782             fjz0             = _mm_add_ps(fjz0,tz);
 783
 784             /**************************
 785              * CALCULATE INTERACTIONS *
 786              **************************/
 787
 788             /* Compute parameters for interactions between i and j atoms */
 789             qq20             = _mm_mul_ps(iq2,jq0);
 790
 791             /* COULOMB ELECTROSTATICS */
 792             velec            = _mm_mul_ps(qq20,rinv20);
 793             felec            = _mm_mul_ps(velec,rinvsq20);
 794
 795             fscal            = felec;
 796
 797             /* Calculate temporary vectorial force */
 798             tx               = _mm_mul_ps(fscal,dx20);
 799             ty               = _mm_mul_ps(fscal,dy20);
 800             tz               = _mm_mul_ps(fscal,dz20);
 801
 802             /* Update vectorial force */
 803             fix2             = _mm_add_ps(fix2,tx);
 804             fiy2             = _mm_add_ps(fiy2,ty);
 805             fiz2             = _mm_add_ps(fiz2,tz);
 806
 807             fjx0             = _mm_add_ps(fjx0,tx);
 808             fjy0             = _mm_add_ps(fjy0,ty);
 809             fjz0             = _mm_add_ps(fjz0,tz);
 810
 811             fjptrA             = f+j_coord_offsetA;
 812             fjptrB             = f+j_coord_offsetB;
 813             fjptrC             = f+j_coord_offsetC;
 814             fjptrD             = f+j_coord_offsetD;
 815
 816             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 817
 818             /* Inner loop uses 88 flops */
 819         }
 820
 821         if(jidx<j_index_end)
 822         {
 823
 824             /* Get j neighbor index, and coordinate index */
 825             jnrlistA         = jjnr[jidx];
 826             jnrlistB         = jjnr[jidx+1];
 827             jnrlistC         = jjnr[jidx+2];
 828             jnrlistD         = jjnr[jidx+3];
 829             /* Sign of each element will be negative for non-real atoms.
 830              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 831              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 832              */
 833             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 834             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 835             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 836             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 837             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 838             j_coord_offsetA  = DIM*jnrA;
 839             j_coord_offsetB  = DIM*jnrB;
 840             j_coord_offsetC  = DIM*jnrC;
 841             j_coord_offsetD  = DIM*jnrD;
 842
 843             /* load j atom coordinates */
 844             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 845                                               x+j_coord_offsetC,x+j_coord_offsetD,
 846                                               &jx0,&jy0,&jz0);
 847
 848             /* Calculate displacement vector */
 849             dx00             = _mm_sub_ps(ix0,jx0);
 850             dy00             = _mm_sub_ps(iy0,jy0);
 851             dz00             = _mm_sub_ps(iz0,jz0);
 852             dx10             = _mm_sub_ps(ix1,jx0);
 853             dy10             = _mm_sub_ps(iy1,jy0);
 854             dz10             = _mm_sub_ps(iz1,jz0);
 855             dx20             = _mm_sub_ps(ix2,jx0);
 856             dy20             = _mm_sub_ps(iy2,jy0);
 857             dz20             = _mm_sub_ps(iz2,jz0);
 858
 859             /* Calculate squared distance and things based on it */
 860             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 861             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 862             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 863
 864             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 865             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 866             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 867
 868             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 869             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 870             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 871
 872             /* Load parameters for j particles */
 873             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 874                                                               charge+jnrC+0,charge+jnrD+0);
 875             vdwjidx0A        = 2*vdwtype[jnrA+0];
 876             vdwjidx0B        = 2*vdwtype[jnrB+0];
 877             vdwjidx0C        = 2*vdwtype[jnrC+0];
 878             vdwjidx0D        = 2*vdwtype[jnrD+0];
 879
 880             fjx0             = _mm_setzero_ps();
 881             fjy0             = _mm_setzero_ps();
 882             fjz0             = _mm_setzero_ps();
 883
 884             /**************************
 885              * CALCULATE INTERACTIONS *
 886              **************************/
 887
 888             /* Compute parameters for interactions between i and j atoms */
 889             qq00             = _mm_mul_ps(iq0,jq0);
 890             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 891                                          vdwparam+vdwioffset0+vdwjidx0B,
 892                                          vdwparam+vdwioffset0+vdwjidx0C,
 893                                          vdwparam+vdwioffset0+vdwjidx0D,
 894                                          &c6_00,&c12_00);
 895
 896             /* COULOMB ELECTROSTATICS */
 897             velec            = _mm_mul_ps(qq00,rinv00);
 898             felec            = _mm_mul_ps(velec,rinvsq00);
 899
 900             /* LENNARD-JONES DISPERSION/REPULSION */
 901
 902             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 903             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 904
 905             fscal            = _mm_add_ps(felec,fvdw);
 906
 907             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 908
 909             /* Calculate temporary vectorial force */
 910             tx               = _mm_mul_ps(fscal,dx00);
 911             ty               = _mm_mul_ps(fscal,dy00);
 912             tz               = _mm_mul_ps(fscal,dz00);
 913
 914             /* Update vectorial force */
 915             fix0             = _mm_add_ps(fix0,tx);
 916             fiy0             = _mm_add_ps(fiy0,ty);
 917             fiz0             = _mm_add_ps(fiz0,tz);
 918
 919             fjx0             = _mm_add_ps(fjx0,tx);
 920             fjy0             = _mm_add_ps(fjy0,ty);
 921             fjz0             = _mm_add_ps(fjz0,tz);
 922
 923             /**************************
 924              * CALCULATE INTERACTIONS *
 925              **************************/
 926
 927             /* Compute parameters for interactions between i and j atoms */
 928             qq10             = _mm_mul_ps(iq1,jq0);
 929
 930             /* COULOMB ELECTROSTATICS */
 931             velec            = _mm_mul_ps(qq10,rinv10);
 932             felec            = _mm_mul_ps(velec,rinvsq10);
 933
 934             fscal            = felec;
 935
 936             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 937
 938             /* Calculate temporary vectorial force */
 939             tx               = _mm_mul_ps(fscal,dx10);
 940             ty               = _mm_mul_ps(fscal,dy10);
 941             tz               = _mm_mul_ps(fscal,dz10);
 942
 943             /* Update vectorial force */
 944             fix1             = _mm_add_ps(fix1,tx);
 945             fiy1             = _mm_add_ps(fiy1,ty);
 946             fiz1             = _mm_add_ps(fiz1,tz);
 947
 948             fjx0             = _mm_add_ps(fjx0,tx);
 949             fjy0             = _mm_add_ps(fjy0,ty);
 950             fjz0             = _mm_add_ps(fjz0,tz);
 951
 952             /**************************
 953              * CALCULATE INTERACTIONS *
 954              **************************/
 955
 956             /* Compute parameters for interactions between i and j atoms */
 957             qq20             = _mm_mul_ps(iq2,jq0);
 958
 959             /* COULOMB ELECTROSTATICS */
 960             velec            = _mm_mul_ps(qq20,rinv20);
 961             felec            = _mm_mul_ps(velec,rinvsq20);
 962
 963             fscal            = felec;
 964
 965             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 966
 967             /* Calculate temporary vectorial force */
 968             tx               = _mm_mul_ps(fscal,dx20);
 969             ty               = _mm_mul_ps(fscal,dy20);
 970             tz               = _mm_mul_ps(fscal,dz20);
 971
 972             /* Update vectorial force */
 973             fix2             = _mm_add_ps(fix2,tx);
 974             fiy2             = _mm_add_ps(fiy2,ty);
 975             fiz2             = _mm_add_ps(fiz2,tz);
 976
 977             fjx0             = _mm_add_ps(fjx0,tx);
 978             fjy0             = _mm_add_ps(fjy0,ty);
 979             fjz0             = _mm_add_ps(fjz0,tz);
 980
 981             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 982             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 983             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 984             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 985
 986             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 987
 988             /* Inner loop uses 88 flops */
 989         }
 990
 991         /* End of innermost loop */
 992
 993         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 994                                               f+i_coord_offset,fshift+i_shift_offset);
 995
 996         /* Increment number of inner iterations */
 997         inneriter                  += j_index_end - j_index_start;
 998
 999         /* Outer loop uses 18 flops */
1000     }
1001
1002     /* Increment number of outer iterations */
1003     outeriter        += nri;
1004
1005     /* Update outer/inner flops */
1006
1007     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*88);
1008 }