src/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecCoul_VdwLJ_GeomW4P1_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_GeomW4P1_VF_sse2_single
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwLJ_GeomW4P1_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              vdwioffset3;
  76     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  77     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  78     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  79     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  80     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  81     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  82     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  83     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  84     real             *charge;
  85     int              nvdwtype;
  86     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  87     int              *vdwtype;
  88     real             *vdwparam;
  89     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  90     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  91     __m128           dummy_mask,cutoff_mask;
  92     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  93     __m128           one     = _mm_set1_ps(1.0);
  94     __m128           two     = _mm_set1_ps(2.0);
  95     x                = xx[0];
  96     f                = ff[0];
  97
  98     nri              = nlist->nri;
  99     iinr             = nlist->iinr;
 100     jindex           = nlist->jindex;
 101     jjnr             = nlist->jjnr;
 102     shiftidx         = nlist->shift;
 103     gid              = nlist->gid;
 104     shiftvec         = fr->shift_vec[0];
 105     fshift           = fr->fshift[0];
 106     facel            = _mm_set1_ps(fr->epsfac);
 107     charge           = mdatoms->chargeA;
 108     nvdwtype         = fr->ntype;
 109     vdwparam         = fr->nbfp;
 110     vdwtype          = mdatoms->typeA;
 111
 112     /* Setup water-specific parameters */
 113     inr              = nlist->iinr[0];
 114     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 115     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 116     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 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_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 150                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 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         fix3             = _mm_setzero_ps();
 162         fiy3             = _mm_setzero_ps();
 163         fiz3             = _mm_setzero_ps();
 164
 165         /* Reset potential sums */
 166         velecsum         = _mm_setzero_ps();
 167         vvdwsum          = _mm_setzero_ps();
 168
 169         /* Start inner kernel loop */
 170         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 171         {
 172
 173             /* Get j neighbor index, and coordinate index */
 174             jnrA             = jjnr[jidx];
 175             jnrB             = jjnr[jidx+1];
 176             jnrC             = jjnr[jidx+2];
 177             jnrD             = jjnr[jidx+3];
 178             j_coord_offsetA  = DIM*jnrA;
 179             j_coord_offsetB  = DIM*jnrB;
 180             j_coord_offsetC  = DIM*jnrC;
 181             j_coord_offsetD  = DIM*jnrD;
 182
 183             /* load j atom coordinates */
 184             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 185                                               x+j_coord_offsetC,x+j_coord_offsetD,
 186                                               &jx0,&jy0,&jz0);
 187
 188             /* Calculate displacement vector */
 189             dx00             = _mm_sub_ps(ix0,jx0);
 190             dy00             = _mm_sub_ps(iy0,jy0);
 191             dz00             = _mm_sub_ps(iz0,jz0);
 192             dx10             = _mm_sub_ps(ix1,jx0);
 193             dy10             = _mm_sub_ps(iy1,jy0);
 194             dz10             = _mm_sub_ps(iz1,jz0);
 195             dx20             = _mm_sub_ps(ix2,jx0);
 196             dy20             = _mm_sub_ps(iy2,jy0);
 197             dz20             = _mm_sub_ps(iz2,jz0);
 198             dx30             = _mm_sub_ps(ix3,jx0);
 199             dy30             = _mm_sub_ps(iy3,jy0);
 200             dz30             = _mm_sub_ps(iz3,jz0);
 201
 202             /* Calculate squared distance and things based on it */
 203             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 204             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 205             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 206             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 207
 208             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 209             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 210             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 211
 212             rinvsq00         = gmx_mm_inv_ps(rsq00);
 213             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 214             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 215             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 216
 217             /* Load parameters for j particles */
 218             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 219                                                               charge+jnrC+0,charge+jnrD+0);
 220             vdwjidx0A        = 2*vdwtype[jnrA+0];
 221             vdwjidx0B        = 2*vdwtype[jnrB+0];
 222             vdwjidx0C        = 2*vdwtype[jnrC+0];
 223             vdwjidx0D        = 2*vdwtype[jnrD+0];
 224
 225             fjx0             = _mm_setzero_ps();
 226             fjy0             = _mm_setzero_ps();
 227             fjz0             = _mm_setzero_ps();
 228
 229             /**************************
 230              * CALCULATE INTERACTIONS *
 231              **************************/
 232
 233             /* Compute parameters for interactions between i and j atoms */
 234             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 235                                          vdwparam+vdwioffset0+vdwjidx0B,
 236                                          vdwparam+vdwioffset0+vdwjidx0C,
 237                                          vdwparam+vdwioffset0+vdwjidx0D,
 238                                          &c6_00,&c12_00);
 239
 240             /* LENNARD-JONES DISPERSION/REPULSION */
 241
 242             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 243             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 244             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 245             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 246             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 247
 248             /* Update potential sum for this i atom from the interaction with this j atom. */
 249             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 250
 251             fscal            = fvdw;
 252
 253             /* Calculate temporary vectorial force */
 254             tx               = _mm_mul_ps(fscal,dx00);
 255             ty               = _mm_mul_ps(fscal,dy00);
 256             tz               = _mm_mul_ps(fscal,dz00);
 257
 258             /* Update vectorial force */
 259             fix0             = _mm_add_ps(fix0,tx);
 260             fiy0             = _mm_add_ps(fiy0,ty);
 261             fiz0             = _mm_add_ps(fiz0,tz);
 262
 263             fjx0             = _mm_add_ps(fjx0,tx);
 264             fjy0             = _mm_add_ps(fjy0,ty);
 265             fjz0             = _mm_add_ps(fjz0,tz);
 266
 267             /**************************
 268              * CALCULATE INTERACTIONS *
 269              **************************/
 270
 271             /* Compute parameters for interactions between i and j atoms */
 272             qq10             = _mm_mul_ps(iq1,jq0);
 273
 274             /* COULOMB ELECTROSTATICS */
 275             velec            = _mm_mul_ps(qq10,rinv10);
 276             felec            = _mm_mul_ps(velec,rinvsq10);
 277
 278             /* Update potential sum for this i atom from the interaction with this j atom. */
 279             velecsum         = _mm_add_ps(velecsum,velec);
 280
 281             fscal            = felec;
 282
 283             /* Calculate temporary vectorial force */
 284             tx               = _mm_mul_ps(fscal,dx10);
 285             ty               = _mm_mul_ps(fscal,dy10);
 286             tz               = _mm_mul_ps(fscal,dz10);
 287
 288             /* Update vectorial force */
 289             fix1             = _mm_add_ps(fix1,tx);
 290             fiy1             = _mm_add_ps(fiy1,ty);
 291             fiz1             = _mm_add_ps(fiz1,tz);
 292
 293             fjx0             = _mm_add_ps(fjx0,tx);
 294             fjy0             = _mm_add_ps(fjy0,ty);
 295             fjz0             = _mm_add_ps(fjz0,tz);
 296
 297             /**************************
 298              * CALCULATE INTERACTIONS *
 299              **************************/
 300
 301             /* Compute parameters for interactions between i and j atoms */
 302             qq20             = _mm_mul_ps(iq2,jq0);
 303
 304             /* COULOMB ELECTROSTATICS */
 305             velec            = _mm_mul_ps(qq20,rinv20);
 306             felec            = _mm_mul_ps(velec,rinvsq20);
 307
 308             /* Update potential sum for this i atom from the interaction with this j atom. */
 309             velecsum         = _mm_add_ps(velecsum,velec);
 310
 311             fscal            = felec;
 312
 313             /* Calculate temporary vectorial force */
 314             tx               = _mm_mul_ps(fscal,dx20);
 315             ty               = _mm_mul_ps(fscal,dy20);
 316             tz               = _mm_mul_ps(fscal,dz20);
 317
 318             /* Update vectorial force */
 319             fix2             = _mm_add_ps(fix2,tx);
 320             fiy2             = _mm_add_ps(fiy2,ty);
 321             fiz2             = _mm_add_ps(fiz2,tz);
 322
 323             fjx0             = _mm_add_ps(fjx0,tx);
 324             fjy0             = _mm_add_ps(fjy0,ty);
 325             fjz0             = _mm_add_ps(fjz0,tz);
 326
 327             /**************************
 328              * CALCULATE INTERACTIONS *
 329              **************************/
 330
 331             /* Compute parameters for interactions between i and j atoms */
 332             qq30             = _mm_mul_ps(iq3,jq0);
 333
 334             /* COULOMB ELECTROSTATICS */
 335             velec            = _mm_mul_ps(qq30,rinv30);
 336             felec            = _mm_mul_ps(velec,rinvsq30);
 337
 338             /* Update potential sum for this i atom from the interaction with this j atom. */
 339             velecsum         = _mm_add_ps(velecsum,velec);
 340
 341             fscal            = felec;
 342
 343             /* Calculate temporary vectorial force */
 344             tx               = _mm_mul_ps(fscal,dx30);
 345             ty               = _mm_mul_ps(fscal,dy30);
 346             tz               = _mm_mul_ps(fscal,dz30);
 347
 348             /* Update vectorial force */
 349             fix3             = _mm_add_ps(fix3,tx);
 350             fiy3             = _mm_add_ps(fiy3,ty);
 351             fiz3             = _mm_add_ps(fiz3,tz);
 352
 353             fjx0             = _mm_add_ps(fjx0,tx);
 354             fjy0             = _mm_add_ps(fjy0,ty);
 355             fjz0             = _mm_add_ps(fjz0,tz);
 356
 357             fjptrA             = f+j_coord_offsetA;
 358             fjptrB             = f+j_coord_offsetB;
 359             fjptrC             = f+j_coord_offsetC;
 360             fjptrD             = f+j_coord_offsetD;
 361
 362             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 363
 364             /* Inner loop uses 116 flops */
 365         }
 366
 367         if(jidx<j_index_end)
 368         {
 369
 370             /* Get j neighbor index, and coordinate index */
 371             jnrlistA         = jjnr[jidx];
 372             jnrlistB         = jjnr[jidx+1];
 373             jnrlistC         = jjnr[jidx+2];
 374             jnrlistD         = jjnr[jidx+3];
 375             /* Sign of each element will be negative for non-real atoms.
 376              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 377              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 378              */
 379             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 380             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 381             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 382             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 383             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 384             j_coord_offsetA  = DIM*jnrA;
 385             j_coord_offsetB  = DIM*jnrB;
 386             j_coord_offsetC  = DIM*jnrC;
 387             j_coord_offsetD  = DIM*jnrD;
 388
 389             /* load j atom coordinates */
 390             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 391                                               x+j_coord_offsetC,x+j_coord_offsetD,
 392                                               &jx0,&jy0,&jz0);
 393
 394             /* Calculate displacement vector */
 395             dx00             = _mm_sub_ps(ix0,jx0);
 396             dy00             = _mm_sub_ps(iy0,jy0);
 397             dz00             = _mm_sub_ps(iz0,jz0);
 398             dx10             = _mm_sub_ps(ix1,jx0);
 399             dy10             = _mm_sub_ps(iy1,jy0);
 400             dz10             = _mm_sub_ps(iz1,jz0);
 401             dx20             = _mm_sub_ps(ix2,jx0);
 402             dy20             = _mm_sub_ps(iy2,jy0);
 403             dz20             = _mm_sub_ps(iz2,jz0);
 404             dx30             = _mm_sub_ps(ix3,jx0);
 405             dy30             = _mm_sub_ps(iy3,jy0);
 406             dz30             = _mm_sub_ps(iz3,jz0);
 407
 408             /* Calculate squared distance and things based on it */
 409             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 410             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 411             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 412             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 413
 414             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 415             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 416             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 417
 418             rinvsq00         = gmx_mm_inv_ps(rsq00);
 419             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 420             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 421             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 422
 423             /* Load parameters for j particles */
 424             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 425                                                               charge+jnrC+0,charge+jnrD+0);
 426             vdwjidx0A        = 2*vdwtype[jnrA+0];
 427             vdwjidx0B        = 2*vdwtype[jnrB+0];
 428             vdwjidx0C        = 2*vdwtype[jnrC+0];
 429             vdwjidx0D        = 2*vdwtype[jnrD+0];
 430
 431             fjx0             = _mm_setzero_ps();
 432             fjy0             = _mm_setzero_ps();
 433             fjz0             = _mm_setzero_ps();
 434
 435             /**************************
 436              * CALCULATE INTERACTIONS *
 437              **************************/
 438
 439             /* Compute parameters for interactions between i and j atoms */
 440             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 441                                          vdwparam+vdwioffset0+vdwjidx0B,
 442                                          vdwparam+vdwioffset0+vdwjidx0C,
 443                                          vdwparam+vdwioffset0+vdwjidx0D,
 444                                          &c6_00,&c12_00);
 445
 446             /* LENNARD-JONES DISPERSION/REPULSION */
 447
 448             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 449             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 450             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 451             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 452             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 453
 454             /* Update potential sum for this i atom from the interaction with this j atom. */
 455             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 456             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 457
 458             fscal            = fvdw;
 459
 460             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 461
 462             /* Calculate temporary vectorial force */
 463             tx               = _mm_mul_ps(fscal,dx00);
 464             ty               = _mm_mul_ps(fscal,dy00);
 465             tz               = _mm_mul_ps(fscal,dz00);
 466
 467             /* Update vectorial force */
 468             fix0             = _mm_add_ps(fix0,tx);
 469             fiy0             = _mm_add_ps(fiy0,ty);
 470             fiz0             = _mm_add_ps(fiz0,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             qq10             = _mm_mul_ps(iq1,jq0);
 482
 483             /* COULOMB ELECTROSTATICS */
 484             velec            = _mm_mul_ps(qq10,rinv10);
 485             felec            = _mm_mul_ps(velec,rinvsq10);
 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,dx10);
 497             ty               = _mm_mul_ps(fscal,dy10);
 498             tz               = _mm_mul_ps(fscal,dz10);
 499
 500             /* Update vectorial force */
 501             fix1             = _mm_add_ps(fix1,tx);
 502             fiy1             = _mm_add_ps(fiy1,ty);
 503             fiz1             = _mm_add_ps(fiz1,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             /**************************
 510              * CALCULATE INTERACTIONS *
 511              **************************/
 512
 513             /* Compute parameters for interactions between i and j atoms */
 514             qq20             = _mm_mul_ps(iq2,jq0);
 515
 516             /* COULOMB ELECTROSTATICS */
 517             velec            = _mm_mul_ps(qq20,rinv20);
 518             felec            = _mm_mul_ps(velec,rinvsq20);
 519
 520             /* Update potential sum for this i atom from the interaction with this j atom. */
 521             velec            = _mm_andnot_ps(dummy_mask,velec);
 522             velecsum         = _mm_add_ps(velecsum,velec);
 523
 524             fscal            = felec;
 525
 526             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 527
 528             /* Calculate temporary vectorial force */
 529             tx               = _mm_mul_ps(fscal,dx20);
 530             ty               = _mm_mul_ps(fscal,dy20);
 531             tz               = _mm_mul_ps(fscal,dz20);
 532
 533             /* Update vectorial force */
 534             fix2             = _mm_add_ps(fix2,tx);
 535             fiy2             = _mm_add_ps(fiy2,ty);
 536             fiz2             = _mm_add_ps(fiz2,tz);
 537
 538             fjx0             = _mm_add_ps(fjx0,tx);
 539             fjy0             = _mm_add_ps(fjy0,ty);
 540             fjz0             = _mm_add_ps(fjz0,tz);
 541
 542             /**************************
 543              * CALCULATE INTERACTIONS *
 544              **************************/
 545
 546             /* Compute parameters for interactions between i and j atoms */
 547             qq30             = _mm_mul_ps(iq3,jq0);
 548
 549             /* COULOMB ELECTROSTATICS */
 550             velec            = _mm_mul_ps(qq30,rinv30);
 551             felec            = _mm_mul_ps(velec,rinvsq30);
 552
 553             /* Update potential sum for this i atom from the interaction with this j atom. */
 554             velec            = _mm_andnot_ps(dummy_mask,velec);
 555             velecsum         = _mm_add_ps(velecsum,velec);
 556
 557             fscal            = felec;
 558
 559             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 560
 561             /* Calculate temporary vectorial force */
 562             tx               = _mm_mul_ps(fscal,dx30);
 563             ty               = _mm_mul_ps(fscal,dy30);
 564             tz               = _mm_mul_ps(fscal,dz30);
 565
 566             /* Update vectorial force */
 567             fix3             = _mm_add_ps(fix3,tx);
 568             fiy3             = _mm_add_ps(fiy3,ty);
 569             fiz3             = _mm_add_ps(fiz3,tz);
 570
 571             fjx0             = _mm_add_ps(fjx0,tx);
 572             fjy0             = _mm_add_ps(fjy0,ty);
 573             fjz0             = _mm_add_ps(fjz0,tz);
 574
 575             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 576             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 577             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 578             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 579
 580             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 581
 582             /* Inner loop uses 116 flops */
 583         }
 584
 585         /* End of innermost loop */
 586
 587         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 588                                               f+i_coord_offset,fshift+i_shift_offset);
 589
 590         ggid                        = gid[iidx];
 591         /* Update potential energies */
 592         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 593         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 594
 595         /* Increment number of inner iterations */
 596         inneriter                  += j_index_end - j_index_start;
 597
 598         /* Outer loop uses 26 flops */
 599     }
 600
 601     /* Increment number of outer iterations */
 602     outeriter        += nri;
 603
 604     /* Update outer/inner flops */
 605
 606     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*116);
 607 }
 608 /*
 609  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW4P1_F_sse2_single
 610  * Electrostatics interaction: Coulomb
 611  * VdW interaction:            LennardJones
 612  * Geometry:                   Water4-Particle
 613  * Calculate force/pot:        Force
 614  */
 615 void
 616 nb_kernel_ElecCoul_VdwLJ_GeomW4P1_F_sse2_single
 617                     (t_nblist * gmx_restrict                nlist,
 618                      rvec * gmx_restrict                    xx,
 619                      rvec * gmx_restrict                    ff,
 620                      t_forcerec * gmx_restrict              fr,
 621                      t_mdatoms * gmx_restrict               mdatoms,
 622                      nb_kernel_data_t * gmx_restrict        kernel_data,
 623                      t_nrnb * gmx_restrict                  nrnb)
 624 {
 625     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 626      * just 0 for non-waters.
 627      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 628      * jnr indices corresponding to data put in the four positions in the SIMD register.
 629      */
 630     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 631     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 632     int              jnrA,jnrB,jnrC,jnrD;
 633     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 634     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 635     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 636     real             rcutoff_scalar;
 637     real             *shiftvec,*fshift,*x,*f;
 638     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 639     real             scratch[4*DIM];
 640     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 641     int              vdwioffset0;
 642     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 643     int              vdwioffset1;
 644     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 645     int              vdwioffset2;
 646     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 647     int              vdwioffset3;
 648     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 649     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 650     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 651     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 652     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 653     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 654     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 655     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 656     real             *charge;
 657     int              nvdwtype;
 658     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 659     int              *vdwtype;
 660     real             *vdwparam;
 661     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 662     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 663     __m128           dummy_mask,cutoff_mask;
 664     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 665     __m128           one     = _mm_set1_ps(1.0);
 666     __m128           two     = _mm_set1_ps(2.0);
 667     x                = xx[0];
 668     f                = ff[0];
 669
 670     nri              = nlist->nri;
 671     iinr             = nlist->iinr;
 672     jindex           = nlist->jindex;
 673     jjnr             = nlist->jjnr;
 674     shiftidx         = nlist->shift;
 675     gid              = nlist->gid;
 676     shiftvec         = fr->shift_vec[0];
 677     fshift           = fr->fshift[0];
 678     facel            = _mm_set1_ps(fr->epsfac);
 679     charge           = mdatoms->chargeA;
 680     nvdwtype         = fr->ntype;
 681     vdwparam         = fr->nbfp;
 682     vdwtype          = mdatoms->typeA;
 683
 684     /* Setup water-specific parameters */
 685     inr              = nlist->iinr[0];
 686     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 687     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 688     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 689     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 690
 691     /* Avoid stupid compiler warnings */
 692     jnrA = jnrB = jnrC = jnrD = 0;
 693     j_coord_offsetA = 0;
 694     j_coord_offsetB = 0;
 695     j_coord_offsetC = 0;
 696     j_coord_offsetD = 0;
 697
 698     outeriter        = 0;
 699     inneriter        = 0;
 700
 701     for(iidx=0;iidx<4*DIM;iidx++)
 702     {
 703         scratch[iidx] = 0.0;
 704     }
 705
 706     /* Start outer loop over neighborlists */
 707     for(iidx=0; iidx<nri; iidx++)
 708     {
 709         /* Load shift vector for this list */
 710         i_shift_offset   = DIM*shiftidx[iidx];
 711
 712         /* Load limits for loop over neighbors */
 713         j_index_start    = jindex[iidx];
 714         j_index_end      = jindex[iidx+1];
 715
 716         /* Get outer coordinate index */
 717         inr              = iinr[iidx];
 718         i_coord_offset   = DIM*inr;
 719
 720         /* Load i particle coords and add shift vector */
 721         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 722                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 723
 724         fix0             = _mm_setzero_ps();
 725         fiy0             = _mm_setzero_ps();
 726         fiz0             = _mm_setzero_ps();
 727         fix1             = _mm_setzero_ps();
 728         fiy1             = _mm_setzero_ps();
 729         fiz1             = _mm_setzero_ps();
 730         fix2             = _mm_setzero_ps();
 731         fiy2             = _mm_setzero_ps();
 732         fiz2             = _mm_setzero_ps();
 733         fix3             = _mm_setzero_ps();
 734         fiy3             = _mm_setzero_ps();
 735         fiz3             = _mm_setzero_ps();
 736
 737         /* Start inner kernel loop */
 738         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 739         {
 740
 741             /* Get j neighbor index, and coordinate index */
 742             jnrA             = jjnr[jidx];
 743             jnrB             = jjnr[jidx+1];
 744             jnrC             = jjnr[jidx+2];
 745             jnrD             = jjnr[jidx+3];
 746             j_coord_offsetA  = DIM*jnrA;
 747             j_coord_offsetB  = DIM*jnrB;
 748             j_coord_offsetC  = DIM*jnrC;
 749             j_coord_offsetD  = DIM*jnrD;
 750
 751             /* load j atom coordinates */
 752             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 753                                               x+j_coord_offsetC,x+j_coord_offsetD,
 754                                               &jx0,&jy0,&jz0);
 755
 756             /* Calculate displacement vector */
 757             dx00             = _mm_sub_ps(ix0,jx0);
 758             dy00             = _mm_sub_ps(iy0,jy0);
 759             dz00             = _mm_sub_ps(iz0,jz0);
 760             dx10             = _mm_sub_ps(ix1,jx0);
 761             dy10             = _mm_sub_ps(iy1,jy0);
 762             dz10             = _mm_sub_ps(iz1,jz0);
 763             dx20             = _mm_sub_ps(ix2,jx0);
 764             dy20             = _mm_sub_ps(iy2,jy0);
 765             dz20             = _mm_sub_ps(iz2,jz0);
 766             dx30             = _mm_sub_ps(ix3,jx0);
 767             dy30             = _mm_sub_ps(iy3,jy0);
 768             dz30             = _mm_sub_ps(iz3,jz0);
 769
 770             /* Calculate squared distance and things based on it */
 771             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 772             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 773             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 774             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 775
 776             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 777             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 778             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 779
 780             rinvsq00         = gmx_mm_inv_ps(rsq00);
 781             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 782             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 783             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 784
 785             /* Load parameters for j particles */
 786             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 787                                                               charge+jnrC+0,charge+jnrD+0);
 788             vdwjidx0A        = 2*vdwtype[jnrA+0];
 789             vdwjidx0B        = 2*vdwtype[jnrB+0];
 790             vdwjidx0C        = 2*vdwtype[jnrC+0];
 791             vdwjidx0D        = 2*vdwtype[jnrD+0];
 792
 793             fjx0             = _mm_setzero_ps();
 794             fjy0             = _mm_setzero_ps();
 795             fjz0             = _mm_setzero_ps();
 796
 797             /**************************
 798              * CALCULATE INTERACTIONS *
 799              **************************/
 800
 801             /* Compute parameters for interactions between i and j atoms */
 802             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 803                                          vdwparam+vdwioffset0+vdwjidx0B,
 804                                          vdwparam+vdwioffset0+vdwjidx0C,
 805                                          vdwparam+vdwioffset0+vdwjidx0D,
 806                                          &c6_00,&c12_00);
 807
 808             /* LENNARD-JONES DISPERSION/REPULSION */
 809
 810             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 811             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 812
 813             fscal            = fvdw;
 814
 815             /* Calculate temporary vectorial force */
 816             tx               = _mm_mul_ps(fscal,dx00);
 817             ty               = _mm_mul_ps(fscal,dy00);
 818             tz               = _mm_mul_ps(fscal,dz00);
 819
 820             /* Update vectorial force */
 821             fix0             = _mm_add_ps(fix0,tx);
 822             fiy0             = _mm_add_ps(fiy0,ty);
 823             fiz0             = _mm_add_ps(fiz0,tz);
 824
 825             fjx0             = _mm_add_ps(fjx0,tx);
 826             fjy0             = _mm_add_ps(fjy0,ty);
 827             fjz0             = _mm_add_ps(fjz0,tz);
 828
 829             /**************************
 830              * CALCULATE INTERACTIONS *
 831              **************************/
 832
 833             /* Compute parameters for interactions between i and j atoms */
 834             qq10             = _mm_mul_ps(iq1,jq0);
 835
 836             /* COULOMB ELECTROSTATICS */
 837             velec            = _mm_mul_ps(qq10,rinv10);
 838             felec            = _mm_mul_ps(velec,rinvsq10);
 839
 840             fscal            = felec;
 841
 842             /* Calculate temporary vectorial force */
 843             tx               = _mm_mul_ps(fscal,dx10);
 844             ty               = _mm_mul_ps(fscal,dy10);
 845             tz               = _mm_mul_ps(fscal,dz10);
 846
 847             /* Update vectorial force */
 848             fix1             = _mm_add_ps(fix1,tx);
 849             fiy1             = _mm_add_ps(fiy1,ty);
 850             fiz1             = _mm_add_ps(fiz1,tz);
 851
 852             fjx0             = _mm_add_ps(fjx0,tx);
 853             fjy0             = _mm_add_ps(fjy0,ty);
 854             fjz0             = _mm_add_ps(fjz0,tz);
 855
 856             /**************************
 857              * CALCULATE INTERACTIONS *
 858              **************************/
 859
 860             /* Compute parameters for interactions between i and j atoms */
 861             qq20             = _mm_mul_ps(iq2,jq0);
 862
 863             /* COULOMB ELECTROSTATICS */
 864             velec            = _mm_mul_ps(qq20,rinv20);
 865             felec            = _mm_mul_ps(velec,rinvsq20);
 866
 867             fscal            = felec;
 868
 869             /* Calculate temporary vectorial force */
 870             tx               = _mm_mul_ps(fscal,dx20);
 871             ty               = _mm_mul_ps(fscal,dy20);
 872             tz               = _mm_mul_ps(fscal,dz20);
 873
 874             /* Update vectorial force */
 875             fix2             = _mm_add_ps(fix2,tx);
 876             fiy2             = _mm_add_ps(fiy2,ty);
 877             fiz2             = _mm_add_ps(fiz2,tz);
 878
 879             fjx0             = _mm_add_ps(fjx0,tx);
 880             fjy0             = _mm_add_ps(fjy0,ty);
 881             fjz0             = _mm_add_ps(fjz0,tz);
 882
 883             /**************************
 884              * CALCULATE INTERACTIONS *
 885              **************************/
 886
 887             /* Compute parameters for interactions between i and j atoms */
 888             qq30             = _mm_mul_ps(iq3,jq0);
 889
 890             /* COULOMB ELECTROSTATICS */
 891             velec            = _mm_mul_ps(qq30,rinv30);
 892             felec            = _mm_mul_ps(velec,rinvsq30);
 893
 894             fscal            = felec;
 895
 896             /* Calculate temporary vectorial force */
 897             tx               = _mm_mul_ps(fscal,dx30);
 898             ty               = _mm_mul_ps(fscal,dy30);
 899             tz               = _mm_mul_ps(fscal,dz30);
 900
 901             /* Update vectorial force */
 902             fix3             = _mm_add_ps(fix3,tx);
 903             fiy3             = _mm_add_ps(fiy3,ty);
 904             fiz3             = _mm_add_ps(fiz3,tz);
 905
 906             fjx0             = _mm_add_ps(fjx0,tx);
 907             fjy0             = _mm_add_ps(fjy0,ty);
 908             fjz0             = _mm_add_ps(fjz0,tz);
 909
 910             fjptrA             = f+j_coord_offsetA;
 911             fjptrB             = f+j_coord_offsetB;
 912             fjptrC             = f+j_coord_offsetC;
 913             fjptrD             = f+j_coord_offsetD;
 914
 915             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 916
 917             /* Inner loop uses 108 flops */
 918         }
 919
 920         if(jidx<j_index_end)
 921         {
 922
 923             /* Get j neighbor index, and coordinate index */
 924             jnrlistA         = jjnr[jidx];
 925             jnrlistB         = jjnr[jidx+1];
 926             jnrlistC         = jjnr[jidx+2];
 927             jnrlistD         = jjnr[jidx+3];
 928             /* Sign of each element will be negative for non-real atoms.
 929              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 930              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 931              */
 932             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 933             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 934             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 935             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 936             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 937             j_coord_offsetA  = DIM*jnrA;
 938             j_coord_offsetB  = DIM*jnrB;
 939             j_coord_offsetC  = DIM*jnrC;
 940             j_coord_offsetD  = DIM*jnrD;
 941
 942             /* load j atom coordinates */
 943             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 944                                               x+j_coord_offsetC,x+j_coord_offsetD,
 945                                               &jx0,&jy0,&jz0);
 946
 947             /* Calculate displacement vector */
 948             dx00             = _mm_sub_ps(ix0,jx0);
 949             dy00             = _mm_sub_ps(iy0,jy0);
 950             dz00             = _mm_sub_ps(iz0,jz0);
 951             dx10             = _mm_sub_ps(ix1,jx0);
 952             dy10             = _mm_sub_ps(iy1,jy0);
 953             dz10             = _mm_sub_ps(iz1,jz0);
 954             dx20             = _mm_sub_ps(ix2,jx0);
 955             dy20             = _mm_sub_ps(iy2,jy0);
 956             dz20             = _mm_sub_ps(iz2,jz0);
 957             dx30             = _mm_sub_ps(ix3,jx0);
 958             dy30             = _mm_sub_ps(iy3,jy0);
 959             dz30             = _mm_sub_ps(iz3,jz0);
 960
 961             /* Calculate squared distance and things based on it */
 962             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 963             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 964             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 965             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 966
 967             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 968             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 969             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 970
 971             rinvsq00         = gmx_mm_inv_ps(rsq00);
 972             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 973             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 974             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 975
 976             /* Load parameters for j particles */
 977             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 978                                                               charge+jnrC+0,charge+jnrD+0);
 979             vdwjidx0A        = 2*vdwtype[jnrA+0];
 980             vdwjidx0B        = 2*vdwtype[jnrB+0];
 981             vdwjidx0C        = 2*vdwtype[jnrC+0];
 982             vdwjidx0D        = 2*vdwtype[jnrD+0];
 983
 984             fjx0             = _mm_setzero_ps();
 985             fjy0             = _mm_setzero_ps();
 986             fjz0             = _mm_setzero_ps();
 987
 988             /**************************
 989              * CALCULATE INTERACTIONS *
 990              **************************/
 991
 992             /* Compute parameters for interactions between i and j atoms */
 993             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 994                                          vdwparam+vdwioffset0+vdwjidx0B,
 995                                          vdwparam+vdwioffset0+vdwjidx0C,
 996                                          vdwparam+vdwioffset0+vdwjidx0D,
 997                                          &c6_00,&c12_00);
 998
 999             /* LENNARD-JONES DISPERSION/REPULSION */
1000
1001             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1002             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1003
1004             fscal            = fvdw;
1005
1006             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1007
1008             /* Calculate temporary vectorial force */
1009             tx               = _mm_mul_ps(fscal,dx00);
1010             ty               = _mm_mul_ps(fscal,dy00);
1011             tz               = _mm_mul_ps(fscal,dz00);
1012
1013             /* Update vectorial force */
1014             fix0             = _mm_add_ps(fix0,tx);
1015             fiy0             = _mm_add_ps(fiy0,ty);
1016             fiz0             = _mm_add_ps(fiz0,tz);
1017
1018             fjx0             = _mm_add_ps(fjx0,tx);
1019             fjy0             = _mm_add_ps(fjy0,ty);
1020             fjz0             = _mm_add_ps(fjz0,tz);
1021
1022             /**************************
1023              * CALCULATE INTERACTIONS *
1024              **************************/
1025
1026             /* Compute parameters for interactions between i and j atoms */
1027             qq10             = _mm_mul_ps(iq1,jq0);
1028
1029             /* COULOMB ELECTROSTATICS */
1030             velec            = _mm_mul_ps(qq10,rinv10);
1031             felec            = _mm_mul_ps(velec,rinvsq10);
1032
1033             fscal            = felec;
1034
1035             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1036
1037             /* Calculate temporary vectorial force */
1038             tx               = _mm_mul_ps(fscal,dx10);
1039             ty               = _mm_mul_ps(fscal,dy10);
1040             tz               = _mm_mul_ps(fscal,dz10);
1041
1042             /* Update vectorial force */
1043             fix1             = _mm_add_ps(fix1,tx);
1044             fiy1             = _mm_add_ps(fiy1,ty);
1045             fiz1             = _mm_add_ps(fiz1,tz);
1046
1047             fjx0             = _mm_add_ps(fjx0,tx);
1048             fjy0             = _mm_add_ps(fjy0,ty);
1049             fjz0             = _mm_add_ps(fjz0,tz);
1050
1051             /**************************
1052              * CALCULATE INTERACTIONS *
1053              **************************/
1054
1055             /* Compute parameters for interactions between i and j atoms */
1056             qq20             = _mm_mul_ps(iq2,jq0);
1057
1058             /* COULOMB ELECTROSTATICS */
1059             velec            = _mm_mul_ps(qq20,rinv20);
1060             felec            = _mm_mul_ps(velec,rinvsq20);
1061
1062             fscal            = felec;
1063
1064             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1065
1066             /* Calculate temporary vectorial force */
1067             tx               = _mm_mul_ps(fscal,dx20);
1068             ty               = _mm_mul_ps(fscal,dy20);
1069             tz               = _mm_mul_ps(fscal,dz20);
1070
1071             /* Update vectorial force */
1072             fix2             = _mm_add_ps(fix2,tx);
1073             fiy2             = _mm_add_ps(fiy2,ty);
1074             fiz2             = _mm_add_ps(fiz2,tz);
1075
1076             fjx0             = _mm_add_ps(fjx0,tx);
1077             fjy0             = _mm_add_ps(fjy0,ty);
1078             fjz0             = _mm_add_ps(fjz0,tz);
1079
1080             /**************************
1081              * CALCULATE INTERACTIONS *
1082              **************************/
1083
1084             /* Compute parameters for interactions between i and j atoms */
1085             qq30             = _mm_mul_ps(iq3,jq0);
1086
1087             /* COULOMB ELECTROSTATICS */
1088             velec            = _mm_mul_ps(qq30,rinv30);
1089             felec            = _mm_mul_ps(velec,rinvsq30);
1090
1091             fscal            = felec;
1092
1093             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1094
1095             /* Calculate temporary vectorial force */
1096             tx               = _mm_mul_ps(fscal,dx30);
1097             ty               = _mm_mul_ps(fscal,dy30);
1098             tz               = _mm_mul_ps(fscal,dz30);
1099
1100             /* Update vectorial force */
1101             fix3             = _mm_add_ps(fix3,tx);
1102             fiy3             = _mm_add_ps(fiy3,ty);
1103             fiz3             = _mm_add_ps(fiz3,tz);
1104
1105             fjx0             = _mm_add_ps(fjx0,tx);
1106             fjy0             = _mm_add_ps(fjy0,ty);
1107             fjz0             = _mm_add_ps(fjz0,tz);
1108
1109             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1110             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1111             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1112             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1113
1114             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1115
1116             /* Inner loop uses 108 flops */
1117         }
1118
1119         /* End of innermost loop */
1120
1121         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1122                                               f+i_coord_offset,fshift+i_shift_offset);
1123
1124         /* Increment number of inner iterations */
1125         inneriter                  += j_index_end - j_index_start;
1126
1127         /* Outer loop uses 24 flops */
1128     }
1129
1130     /* Increment number of outer iterations */
1131     outeriter        += nri;
1132
1133     /* Update outer/inner flops */
1134
1135     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*108);
1136 }