src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRF_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_ElecRF_VdwLJ_GeomW3P1_VF_sse2_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRF_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     krf              = _mm_set1_ps(fr->ic->k_rf);
 106     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 107     crf              = _mm_set1_ps(fr->ic->c_rf);
 108     nvdwtype         = fr->ntype;
 109     vdwparam         = fr->nbfp;
 110     vdwtype          = mdatoms->typeA;
 111
 112     /* Setup water-specific parameters */
 113     inr              = nlist->iinr[0];
 114     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 115     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 116     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 117     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 118
 119     /* Avoid stupid compiler warnings */
 120     jnrA = jnrB = jnrC = jnrD = 0;
 121     j_coord_offsetA = 0;
 122     j_coord_offsetB = 0;
 123     j_coord_offsetC = 0;
 124     j_coord_offsetD = 0;
 125
 126     outeriter        = 0;
 127     inneriter        = 0;
 128
 129     for(iidx=0;iidx<4*DIM;iidx++)
 130     {
 131         scratch[iidx] = 0.0;
 132     }
 133
 134     /* Start outer loop over neighborlists */
 135     for(iidx=0; iidx<nri; iidx++)
 136     {
 137         /* Load shift vector for this list */
 138         i_shift_offset   = DIM*shiftidx[iidx];
 139
 140         /* Load limits for loop over neighbors */
 141         j_index_start    = jindex[iidx];
 142         j_index_end      = jindex[iidx+1];
 143
 144         /* Get outer coordinate index */
 145         inr              = iinr[iidx];
 146         i_coord_offset   = DIM*inr;
 147
 148         /* Load i particle coords and add shift vector */
 149         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 150                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 151
 152         fix0             = _mm_setzero_ps();
 153         fiy0             = _mm_setzero_ps();
 154         fiz0             = _mm_setzero_ps();
 155         fix1             = _mm_setzero_ps();
 156         fiy1             = _mm_setzero_ps();
 157         fiz1             = _mm_setzero_ps();
 158         fix2             = _mm_setzero_ps();
 159         fiy2             = _mm_setzero_ps();
 160         fiz2             = _mm_setzero_ps();
 161
 162         /* Reset potential sums */
 163         velecsum         = _mm_setzero_ps();
 164         vvdwsum          = _mm_setzero_ps();
 165
 166         /* Start inner kernel loop */
 167         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 168         {
 169
 170             /* Get j neighbor index, and coordinate index */
 171             jnrA             = jjnr[jidx];
 172             jnrB             = jjnr[jidx+1];
 173             jnrC             = jjnr[jidx+2];
 174             jnrD             = jjnr[jidx+3];
 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             fjx0             = _mm_setzero_ps();
 218             fjy0             = _mm_setzero_ps();
 219             fjz0             = _mm_setzero_ps();
 220
 221             /**************************
 222              * CALCULATE INTERACTIONS *
 223              **************************/
 224
 225             /* Compute parameters for interactions between i and j atoms */
 226             qq00             = _mm_mul_ps(iq0,jq0);
 227             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 228                                          vdwparam+vdwioffset0+vdwjidx0B,
 229                                          vdwparam+vdwioffset0+vdwjidx0C,
 230                                          vdwparam+vdwioffset0+vdwjidx0D,
 231                                          &c6_00,&c12_00);
 232
 233             /* REACTION-FIELD ELECTROSTATICS */
 234             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 235             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 236
 237             /* LENNARD-JONES DISPERSION/REPULSION */
 238
 239             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 240             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 241             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 242             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 243             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 244
 245             /* Update potential sum for this i atom from the interaction with this j atom. */
 246             velecsum         = _mm_add_ps(velecsum,velec);
 247             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 248
 249             fscal            = _mm_add_ps(felec,fvdw);
 250
 251             /* Calculate temporary vectorial force */
 252             tx               = _mm_mul_ps(fscal,dx00);
 253             ty               = _mm_mul_ps(fscal,dy00);
 254             tz               = _mm_mul_ps(fscal,dz00);
 255
 256             /* Update vectorial force */
 257             fix0             = _mm_add_ps(fix0,tx);
 258             fiy0             = _mm_add_ps(fiy0,ty);
 259             fiz0             = _mm_add_ps(fiz0,tz);
 260
 261             fjx0             = _mm_add_ps(fjx0,tx);
 262             fjy0             = _mm_add_ps(fjy0,ty);
 263             fjz0             = _mm_add_ps(fjz0,tz);
 264
 265             /**************************
 266              * CALCULATE INTERACTIONS *
 267              **************************/
 268
 269             /* Compute parameters for interactions between i and j atoms */
 270             qq10             = _mm_mul_ps(iq1,jq0);
 271
 272             /* REACTION-FIELD ELECTROSTATICS */
 273             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 274             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 275
 276             /* Update potential sum for this i atom from the interaction with this j atom. */
 277             velecsum         = _mm_add_ps(velecsum,velec);
 278
 279             fscal            = felec;
 280
 281             /* Calculate temporary vectorial force */
 282             tx               = _mm_mul_ps(fscal,dx10);
 283             ty               = _mm_mul_ps(fscal,dy10);
 284             tz               = _mm_mul_ps(fscal,dz10);
 285
 286             /* Update vectorial force */
 287             fix1             = _mm_add_ps(fix1,tx);
 288             fiy1             = _mm_add_ps(fiy1,ty);
 289             fiz1             = _mm_add_ps(fiz1,tz);
 290
 291             fjx0             = _mm_add_ps(fjx0,tx);
 292             fjy0             = _mm_add_ps(fjy0,ty);
 293             fjz0             = _mm_add_ps(fjz0,tz);
 294
 295             /**************************
 296              * CALCULATE INTERACTIONS *
 297              **************************/
 298
 299             /* Compute parameters for interactions between i and j atoms */
 300             qq20             = _mm_mul_ps(iq2,jq0);
 301
 302             /* REACTION-FIELD ELECTROSTATICS */
 303             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 304             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 305
 306             /* Update potential sum for this i atom from the interaction with this j atom. */
 307             velecsum         = _mm_add_ps(velecsum,velec);
 308
 309             fscal            = felec;
 310
 311             /* Calculate temporary vectorial force */
 312             tx               = _mm_mul_ps(fscal,dx20);
 313             ty               = _mm_mul_ps(fscal,dy20);
 314             tz               = _mm_mul_ps(fscal,dz20);
 315
 316             /* Update vectorial force */
 317             fix2             = _mm_add_ps(fix2,tx);
 318             fiy2             = _mm_add_ps(fiy2,ty);
 319             fiz2             = _mm_add_ps(fiz2,tz);
 320
 321             fjx0             = _mm_add_ps(fjx0,tx);
 322             fjy0             = _mm_add_ps(fjy0,ty);
 323             fjz0             = _mm_add_ps(fjz0,tz);
 324
 325             fjptrA             = f+j_coord_offsetA;
 326             fjptrB             = f+j_coord_offsetB;
 327             fjptrC             = f+j_coord_offsetC;
 328             fjptrD             = f+j_coord_offsetD;
 329
 330             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 331
 332             /* Inner loop uses 108 flops */
 333         }
 334
 335         if(jidx<j_index_end)
 336         {
 337
 338             /* Get j neighbor index, and coordinate index */
 339             jnrlistA         = jjnr[jidx];
 340             jnrlistB         = jjnr[jidx+1];
 341             jnrlistC         = jjnr[jidx+2];
 342             jnrlistD         = jjnr[jidx+3];
 343             /* Sign of each element will be negative for non-real atoms.
 344              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 345              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 346              */
 347             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 348             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 349             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 350             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 351             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 352             j_coord_offsetA  = DIM*jnrA;
 353             j_coord_offsetB  = DIM*jnrB;
 354             j_coord_offsetC  = DIM*jnrC;
 355             j_coord_offsetD  = DIM*jnrD;
 356
 357             /* load j atom coordinates */
 358             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 359                                               x+j_coord_offsetC,x+j_coord_offsetD,
 360                                               &jx0,&jy0,&jz0);
 361
 362             /* Calculate displacement vector */
 363             dx00             = _mm_sub_ps(ix0,jx0);
 364             dy00             = _mm_sub_ps(iy0,jy0);
 365             dz00             = _mm_sub_ps(iz0,jz0);
 366             dx10             = _mm_sub_ps(ix1,jx0);
 367             dy10             = _mm_sub_ps(iy1,jy0);
 368             dz10             = _mm_sub_ps(iz1,jz0);
 369             dx20             = _mm_sub_ps(ix2,jx0);
 370             dy20             = _mm_sub_ps(iy2,jy0);
 371             dz20             = _mm_sub_ps(iz2,jz0);
 372
 373             /* Calculate squared distance and things based on it */
 374             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 375             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 376             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 377
 378             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 379             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 380             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 381
 382             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 383             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 384             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 385
 386             /* Load parameters for j particles */
 387             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 388                                                               charge+jnrC+0,charge+jnrD+0);
 389             vdwjidx0A        = 2*vdwtype[jnrA+0];
 390             vdwjidx0B        = 2*vdwtype[jnrB+0];
 391             vdwjidx0C        = 2*vdwtype[jnrC+0];
 392             vdwjidx0D        = 2*vdwtype[jnrD+0];
 393
 394             fjx0             = _mm_setzero_ps();
 395             fjy0             = _mm_setzero_ps();
 396             fjz0             = _mm_setzero_ps();
 397
 398             /**************************
 399              * CALCULATE INTERACTIONS *
 400              **************************/
 401
 402             /* Compute parameters for interactions between i and j atoms */
 403             qq00             = _mm_mul_ps(iq0,jq0);
 404             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 405                                          vdwparam+vdwioffset0+vdwjidx0B,
 406                                          vdwparam+vdwioffset0+vdwjidx0C,
 407                                          vdwparam+vdwioffset0+vdwjidx0D,
 408                                          &c6_00,&c12_00);
 409
 410             /* REACTION-FIELD ELECTROSTATICS */
 411             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 412             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 413
 414             /* LENNARD-JONES DISPERSION/REPULSION */
 415
 416             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 417             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 418             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 419             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 420             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 421
 422             /* Update potential sum for this i atom from the interaction with this j atom. */
 423             velec            = _mm_andnot_ps(dummy_mask,velec);
 424             velecsum         = _mm_add_ps(velecsum,velec);
 425             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 426             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 427
 428             fscal            = _mm_add_ps(felec,fvdw);
 429
 430             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 431
 432             /* Calculate temporary vectorial force */
 433             tx               = _mm_mul_ps(fscal,dx00);
 434             ty               = _mm_mul_ps(fscal,dy00);
 435             tz               = _mm_mul_ps(fscal,dz00);
 436
 437             /* Update vectorial force */
 438             fix0             = _mm_add_ps(fix0,tx);
 439             fiy0             = _mm_add_ps(fiy0,ty);
 440             fiz0             = _mm_add_ps(fiz0,tz);
 441
 442             fjx0             = _mm_add_ps(fjx0,tx);
 443             fjy0             = _mm_add_ps(fjy0,ty);
 444             fjz0             = _mm_add_ps(fjz0,tz);
 445
 446             /**************************
 447              * CALCULATE INTERACTIONS *
 448              **************************/
 449
 450             /* Compute parameters for interactions between i and j atoms */
 451             qq10             = _mm_mul_ps(iq1,jq0);
 452
 453             /* REACTION-FIELD ELECTROSTATICS */
 454             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 455             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 456
 457             /* Update potential sum for this i atom from the interaction with this j atom. */
 458             velec            = _mm_andnot_ps(dummy_mask,velec);
 459             velecsum         = _mm_add_ps(velecsum,velec);
 460
 461             fscal            = felec;
 462
 463             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 464
 465             /* Calculate temporary vectorial force */
 466             tx               = _mm_mul_ps(fscal,dx10);
 467             ty               = _mm_mul_ps(fscal,dy10);
 468             tz               = _mm_mul_ps(fscal,dz10);
 469
 470             /* Update vectorial force */
 471             fix1             = _mm_add_ps(fix1,tx);
 472             fiy1             = _mm_add_ps(fiy1,ty);
 473             fiz1             = _mm_add_ps(fiz1,tz);
 474
 475             fjx0             = _mm_add_ps(fjx0,tx);
 476             fjy0             = _mm_add_ps(fjy0,ty);
 477             fjz0             = _mm_add_ps(fjz0,tz);
 478
 479             /**************************
 480              * CALCULATE INTERACTIONS *
 481              **************************/
 482
 483             /* Compute parameters for interactions between i and j atoms */
 484             qq20             = _mm_mul_ps(iq2,jq0);
 485
 486             /* REACTION-FIELD ELECTROSTATICS */
 487             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 488             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 489
 490             /* Update potential sum for this i atom from the interaction with this j atom. */
 491             velec            = _mm_andnot_ps(dummy_mask,velec);
 492             velecsum         = _mm_add_ps(velecsum,velec);
 493
 494             fscal            = felec;
 495
 496             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 497
 498             /* Calculate temporary vectorial force */
 499             tx               = _mm_mul_ps(fscal,dx20);
 500             ty               = _mm_mul_ps(fscal,dy20);
 501             tz               = _mm_mul_ps(fscal,dz20);
 502
 503             /* Update vectorial force */
 504             fix2             = _mm_add_ps(fix2,tx);
 505             fiy2             = _mm_add_ps(fiy2,ty);
 506             fiz2             = _mm_add_ps(fiz2,tz);
 507
 508             fjx0             = _mm_add_ps(fjx0,tx);
 509             fjy0             = _mm_add_ps(fjy0,ty);
 510             fjz0             = _mm_add_ps(fjz0,tz);
 511
 512             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 513             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 514             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 515             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 516
 517             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 518
 519             /* Inner loop uses 108 flops */
 520         }
 521
 522         /* End of innermost loop */
 523
 524         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 525                                               f+i_coord_offset,fshift+i_shift_offset);
 526
 527         ggid                        = gid[iidx];
 528         /* Update potential energies */
 529         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 530         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 531
 532         /* Increment number of inner iterations */
 533         inneriter                  += j_index_end - j_index_start;
 534
 535         /* Outer loop uses 20 flops */
 536     }
 537
 538     /* Increment number of outer iterations */
 539     outeriter        += nri;
 540
 541     /* Update outer/inner flops */
 542
 543     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*108);
 544 }
 545 /*
 546  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomW3P1_F_sse2_single
 547  * Electrostatics interaction: ReactionField
 548  * VdW interaction:            LennardJones
 549  * Geometry:                   Water3-Particle
 550  * Calculate force/pot:        Force
 551  */
 552 void
 553 nb_kernel_ElecRF_VdwLJ_GeomW3P1_F_sse2_single
 554                     (t_nblist * gmx_restrict                nlist,
 555                      rvec * gmx_restrict                    xx,
 556                      rvec * gmx_restrict                    ff,
 557                      t_forcerec * gmx_restrict              fr,
 558                      t_mdatoms * gmx_restrict               mdatoms,
 559                      nb_kernel_data_t * gmx_restrict        kernel_data,
 560                      t_nrnb * gmx_restrict                  nrnb)
 561 {
 562     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 563      * just 0 for non-waters.
 564      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 565      * jnr indices corresponding to data put in the four positions in the SIMD register.
 566      */
 567     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 568     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 569     int              jnrA,jnrB,jnrC,jnrD;
 570     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 571     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 572     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 573     real             rcutoff_scalar;
 574     real             *shiftvec,*fshift,*x,*f;
 575     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 576     real             scratch[4*DIM];
 577     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 578     int              vdwioffset0;
 579     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 580     int              vdwioffset1;
 581     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 582     int              vdwioffset2;
 583     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 584     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 585     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 586     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 587     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 588     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 589     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 590     real             *charge;
 591     int              nvdwtype;
 592     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 593     int              *vdwtype;
 594     real             *vdwparam;
 595     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 596     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 597     __m128           dummy_mask,cutoff_mask;
 598     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 599     __m128           one     = _mm_set1_ps(1.0);
 600     __m128           two     = _mm_set1_ps(2.0);
 601     x                = xx[0];
 602     f                = ff[0];
 603
 604     nri              = nlist->nri;
 605     iinr             = nlist->iinr;
 606     jindex           = nlist->jindex;
 607     jjnr             = nlist->jjnr;
 608     shiftidx         = nlist->shift;
 609     gid              = nlist->gid;
 610     shiftvec         = fr->shift_vec[0];
 611     fshift           = fr->fshift[0];
 612     facel            = _mm_set1_ps(fr->epsfac);
 613     charge           = mdatoms->chargeA;
 614     krf              = _mm_set1_ps(fr->ic->k_rf);
 615     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 616     crf              = _mm_set1_ps(fr->ic->c_rf);
 617     nvdwtype         = fr->ntype;
 618     vdwparam         = fr->nbfp;
 619     vdwtype          = mdatoms->typeA;
 620
 621     /* Setup water-specific parameters */
 622     inr              = nlist->iinr[0];
 623     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 624     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 625     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 626     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 627
 628     /* Avoid stupid compiler warnings */
 629     jnrA = jnrB = jnrC = jnrD = 0;
 630     j_coord_offsetA = 0;
 631     j_coord_offsetB = 0;
 632     j_coord_offsetC = 0;
 633     j_coord_offsetD = 0;
 634
 635     outeriter        = 0;
 636     inneriter        = 0;
 637
 638     for(iidx=0;iidx<4*DIM;iidx++)
 639     {
 640         scratch[iidx] = 0.0;
 641     }
 642
 643     /* Start outer loop over neighborlists */
 644     for(iidx=0; iidx<nri; iidx++)
 645     {
 646         /* Load shift vector for this list */
 647         i_shift_offset   = DIM*shiftidx[iidx];
 648
 649         /* Load limits for loop over neighbors */
 650         j_index_start    = jindex[iidx];
 651         j_index_end      = jindex[iidx+1];
 652
 653         /* Get outer coordinate index */
 654         inr              = iinr[iidx];
 655         i_coord_offset   = DIM*inr;
 656
 657         /* Load i particle coords and add shift vector */
 658         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 659                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 660
 661         fix0             = _mm_setzero_ps();
 662         fiy0             = _mm_setzero_ps();
 663         fiz0             = _mm_setzero_ps();
 664         fix1             = _mm_setzero_ps();
 665         fiy1             = _mm_setzero_ps();
 666         fiz1             = _mm_setzero_ps();
 667         fix2             = _mm_setzero_ps();
 668         fiy2             = _mm_setzero_ps();
 669         fiz2             = _mm_setzero_ps();
 670
 671         /* Start inner kernel loop */
 672         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 673         {
 674
 675             /* Get j neighbor index, and coordinate index */
 676             jnrA             = jjnr[jidx];
 677             jnrB             = jjnr[jidx+1];
 678             jnrC             = jjnr[jidx+2];
 679             jnrD             = jjnr[jidx+3];
 680             j_coord_offsetA  = DIM*jnrA;
 681             j_coord_offsetB  = DIM*jnrB;
 682             j_coord_offsetC  = DIM*jnrC;
 683             j_coord_offsetD  = DIM*jnrD;
 684
 685             /* load j atom coordinates */
 686             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 687                                               x+j_coord_offsetC,x+j_coord_offsetD,
 688                                               &jx0,&jy0,&jz0);
 689
 690             /* Calculate displacement vector */
 691             dx00             = _mm_sub_ps(ix0,jx0);
 692             dy00             = _mm_sub_ps(iy0,jy0);
 693             dz00             = _mm_sub_ps(iz0,jz0);
 694             dx10             = _mm_sub_ps(ix1,jx0);
 695             dy10             = _mm_sub_ps(iy1,jy0);
 696             dz10             = _mm_sub_ps(iz1,jz0);
 697             dx20             = _mm_sub_ps(ix2,jx0);
 698             dy20             = _mm_sub_ps(iy2,jy0);
 699             dz20             = _mm_sub_ps(iz2,jz0);
 700
 701             /* Calculate squared distance and things based on it */
 702             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 703             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 704             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 705
 706             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 707             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 708             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 709
 710             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 711             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 712             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 713
 714             /* Load parameters for j particles */
 715             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 716                                                               charge+jnrC+0,charge+jnrD+0);
 717             vdwjidx0A        = 2*vdwtype[jnrA+0];
 718             vdwjidx0B        = 2*vdwtype[jnrB+0];
 719             vdwjidx0C        = 2*vdwtype[jnrC+0];
 720             vdwjidx0D        = 2*vdwtype[jnrD+0];
 721
 722             fjx0             = _mm_setzero_ps();
 723             fjy0             = _mm_setzero_ps();
 724             fjz0             = _mm_setzero_ps();
 725
 726             /**************************
 727              * CALCULATE INTERACTIONS *
 728              **************************/
 729
 730             /* Compute parameters for interactions between i and j atoms */
 731             qq00             = _mm_mul_ps(iq0,jq0);
 732             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 733                                          vdwparam+vdwioffset0+vdwjidx0B,
 734                                          vdwparam+vdwioffset0+vdwjidx0C,
 735                                          vdwparam+vdwioffset0+vdwjidx0D,
 736                                          &c6_00,&c12_00);
 737
 738             /* REACTION-FIELD ELECTROSTATICS */
 739             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 740
 741             /* LENNARD-JONES DISPERSION/REPULSION */
 742
 743             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 744             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 745
 746             fscal            = _mm_add_ps(felec,fvdw);
 747
 748             /* Calculate temporary vectorial force */
 749             tx               = _mm_mul_ps(fscal,dx00);
 750             ty               = _mm_mul_ps(fscal,dy00);
 751             tz               = _mm_mul_ps(fscal,dz00);
 752
 753             /* Update vectorial force */
 754             fix0             = _mm_add_ps(fix0,tx);
 755             fiy0             = _mm_add_ps(fiy0,ty);
 756             fiz0             = _mm_add_ps(fiz0,tz);
 757
 758             fjx0             = _mm_add_ps(fjx0,tx);
 759             fjy0             = _mm_add_ps(fjy0,ty);
 760             fjz0             = _mm_add_ps(fjz0,tz);
 761
 762             /**************************
 763              * CALCULATE INTERACTIONS *
 764              **************************/
 765
 766             /* Compute parameters for interactions between i and j atoms */
 767             qq10             = _mm_mul_ps(iq1,jq0);
 768
 769             /* REACTION-FIELD ELECTROSTATICS */
 770             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 771
 772             fscal            = felec;
 773
 774             /* Calculate temporary vectorial force */
 775             tx               = _mm_mul_ps(fscal,dx10);
 776             ty               = _mm_mul_ps(fscal,dy10);
 777             tz               = _mm_mul_ps(fscal,dz10);
 778
 779             /* Update vectorial force */
 780             fix1             = _mm_add_ps(fix1,tx);
 781             fiy1             = _mm_add_ps(fiy1,ty);
 782             fiz1             = _mm_add_ps(fiz1,tz);
 783
 784             fjx0             = _mm_add_ps(fjx0,tx);
 785             fjy0             = _mm_add_ps(fjy0,ty);
 786             fjz0             = _mm_add_ps(fjz0,tz);
 787
 788             /**************************
 789              * CALCULATE INTERACTIONS *
 790              **************************/
 791
 792             /* Compute parameters for interactions between i and j atoms */
 793             qq20             = _mm_mul_ps(iq2,jq0);
 794
 795             /* REACTION-FIELD ELECTROSTATICS */
 796             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 797
 798             fscal            = felec;
 799
 800             /* Calculate temporary vectorial force */
 801             tx               = _mm_mul_ps(fscal,dx20);
 802             ty               = _mm_mul_ps(fscal,dy20);
 803             tz               = _mm_mul_ps(fscal,dz20);
 804
 805             /* Update vectorial force */
 806             fix2             = _mm_add_ps(fix2,tx);
 807             fiy2             = _mm_add_ps(fiy2,ty);
 808             fiz2             = _mm_add_ps(fiz2,tz);
 809
 810             fjx0             = _mm_add_ps(fjx0,tx);
 811             fjy0             = _mm_add_ps(fjy0,ty);
 812             fjz0             = _mm_add_ps(fjz0,tz);
 813
 814             fjptrA             = f+j_coord_offsetA;
 815             fjptrB             = f+j_coord_offsetB;
 816             fjptrC             = f+j_coord_offsetC;
 817             fjptrD             = f+j_coord_offsetD;
 818
 819             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 820
 821             /* Inner loop uses 88 flops */
 822         }
 823
 824         if(jidx<j_index_end)
 825         {
 826
 827             /* Get j neighbor index, and coordinate index */
 828             jnrlistA         = jjnr[jidx];
 829             jnrlistB         = jjnr[jidx+1];
 830             jnrlistC         = jjnr[jidx+2];
 831             jnrlistD         = jjnr[jidx+3];
 832             /* Sign of each element will be negative for non-real atoms.
 833              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 834              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 835              */
 836             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 837             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 838             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 839             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 840             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 841             j_coord_offsetA  = DIM*jnrA;
 842             j_coord_offsetB  = DIM*jnrB;
 843             j_coord_offsetC  = DIM*jnrC;
 844             j_coord_offsetD  = DIM*jnrD;
 845
 846             /* load j atom coordinates */
 847             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 848                                               x+j_coord_offsetC,x+j_coord_offsetD,
 849                                               &jx0,&jy0,&jz0);
 850
 851             /* Calculate displacement vector */
 852             dx00             = _mm_sub_ps(ix0,jx0);
 853             dy00             = _mm_sub_ps(iy0,jy0);
 854             dz00             = _mm_sub_ps(iz0,jz0);
 855             dx10             = _mm_sub_ps(ix1,jx0);
 856             dy10             = _mm_sub_ps(iy1,jy0);
 857             dz10             = _mm_sub_ps(iz1,jz0);
 858             dx20             = _mm_sub_ps(ix2,jx0);
 859             dy20             = _mm_sub_ps(iy2,jy0);
 860             dz20             = _mm_sub_ps(iz2,jz0);
 861
 862             /* Calculate squared distance and things based on it */
 863             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 864             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 865             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 866
 867             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 868             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 869             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 870
 871             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 872             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 873             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 874
 875             /* Load parameters for j particles */
 876             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 877                                                               charge+jnrC+0,charge+jnrD+0);
 878             vdwjidx0A        = 2*vdwtype[jnrA+0];
 879             vdwjidx0B        = 2*vdwtype[jnrB+0];
 880             vdwjidx0C        = 2*vdwtype[jnrC+0];
 881             vdwjidx0D        = 2*vdwtype[jnrD+0];
 882
 883             fjx0             = _mm_setzero_ps();
 884             fjy0             = _mm_setzero_ps();
 885             fjz0             = _mm_setzero_ps();
 886
 887             /**************************
 888              * CALCULATE INTERACTIONS *
 889              **************************/
 890
 891             /* Compute parameters for interactions between i and j atoms */
 892             qq00             = _mm_mul_ps(iq0,jq0);
 893             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 894                                          vdwparam+vdwioffset0+vdwjidx0B,
 895                                          vdwparam+vdwioffset0+vdwjidx0C,
 896                                          vdwparam+vdwioffset0+vdwjidx0D,
 897                                          &c6_00,&c12_00);
 898
 899             /* REACTION-FIELD ELECTROSTATICS */
 900             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 901
 902             /* LENNARD-JONES DISPERSION/REPULSION */
 903
 904             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 905             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 906
 907             fscal            = _mm_add_ps(felec,fvdw);
 908
 909             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 910
 911             /* Calculate temporary vectorial force */
 912             tx               = _mm_mul_ps(fscal,dx00);
 913             ty               = _mm_mul_ps(fscal,dy00);
 914             tz               = _mm_mul_ps(fscal,dz00);
 915
 916             /* Update vectorial force */
 917             fix0             = _mm_add_ps(fix0,tx);
 918             fiy0             = _mm_add_ps(fiy0,ty);
 919             fiz0             = _mm_add_ps(fiz0,tz);
 920
 921             fjx0             = _mm_add_ps(fjx0,tx);
 922             fjy0             = _mm_add_ps(fjy0,ty);
 923             fjz0             = _mm_add_ps(fjz0,tz);
 924
 925             /**************************
 926              * CALCULATE INTERACTIONS *
 927              **************************/
 928
 929             /* Compute parameters for interactions between i and j atoms */
 930             qq10             = _mm_mul_ps(iq1,jq0);
 931
 932             /* REACTION-FIELD ELECTROSTATICS */
 933             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 934
 935             fscal            = felec;
 936
 937             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 938
 939             /* Calculate temporary vectorial force */
 940             tx               = _mm_mul_ps(fscal,dx10);
 941             ty               = _mm_mul_ps(fscal,dy10);
 942             tz               = _mm_mul_ps(fscal,dz10);
 943
 944             /* Update vectorial force */
 945             fix1             = _mm_add_ps(fix1,tx);
 946             fiy1             = _mm_add_ps(fiy1,ty);
 947             fiz1             = _mm_add_ps(fiz1,tz);
 948
 949             fjx0             = _mm_add_ps(fjx0,tx);
 950             fjy0             = _mm_add_ps(fjy0,ty);
 951             fjz0             = _mm_add_ps(fjz0,tz);
 952
 953             /**************************
 954              * CALCULATE INTERACTIONS *
 955              **************************/
 956
 957             /* Compute parameters for interactions between i and j atoms */
 958             qq20             = _mm_mul_ps(iq2,jq0);
 959
 960             /* REACTION-FIELD ELECTROSTATICS */
 961             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 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 }