src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_sse2_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse2_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_VF_sse2_single
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_VF_sse2_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  89     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 100     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 101     __m128           dummy_mask,cutoff_mask;
 102     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 103     __m128           one     = _mm_set1_ps(1.0);
 104     __m128           two     = _mm_set1_ps(2.0);
 105     x                = xx[0];
 106     f                = ff[0];
 107
 108     nri              = nlist->nri;
 109     iinr             = nlist->iinr;
 110     jindex           = nlist->jindex;
 111     jjnr             = nlist->jjnr;
 112     shiftidx         = nlist->shift;
 113     gid              = nlist->gid;
 114     shiftvec         = fr->shift_vec[0];
 115     fshift           = fr->fshift[0];
 116     facel            = _mm_set1_ps(fr->ic->epsfac);
 117     charge           = mdatoms->chargeA;
 118     krf              = _mm_set1_ps(fr->ic->k_rf);
 119     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 120     crf              = _mm_set1_ps(fr->ic->c_rf);
 121     nvdwtype         = fr->ntype;
 122     vdwparam         = fr->nbfp;
 123     vdwtype          = mdatoms->typeA;
 124
 125     /* Setup water-specific parameters */
 126     inr              = nlist->iinr[0];
 127     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 128     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 129     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 130     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 131
 132     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 133     rcutoff_scalar   = fr->ic->rcoulomb;
 134     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 135     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 136
 137     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 138     rvdw             = _mm_set1_ps(fr->ic->rvdw);
 139
 140     /* Avoid stupid compiler warnings */
 141     jnrA = jnrB = jnrC = jnrD = 0;
 142     j_coord_offsetA = 0;
 143     j_coord_offsetB = 0;
 144     j_coord_offsetC = 0;
 145     j_coord_offsetD = 0;
 146
 147     outeriter        = 0;
 148     inneriter        = 0;
 149
 150     for(iidx=0;iidx<4*DIM;iidx++)
 151     {
 152         scratch[iidx] = 0.0;
 153     }
 154
 155     /* Start outer loop over neighborlists */
 156     for(iidx=0; iidx<nri; iidx++)
 157     {
 158         /* Load shift vector for this list */
 159         i_shift_offset   = DIM*shiftidx[iidx];
 160
 161         /* Load limits for loop over neighbors */
 162         j_index_start    = jindex[iidx];
 163         j_index_end      = jindex[iidx+1];
 164
 165         /* Get outer coordinate index */
 166         inr              = iinr[iidx];
 167         i_coord_offset   = DIM*inr;
 168
 169         /* Load i particle coords and add shift vector */
 170         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 171                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 172
 173         fix0             = _mm_setzero_ps();
 174         fiy0             = _mm_setzero_ps();
 175         fiz0             = _mm_setzero_ps();
 176         fix1             = _mm_setzero_ps();
 177         fiy1             = _mm_setzero_ps();
 178         fiz1             = _mm_setzero_ps();
 179         fix2             = _mm_setzero_ps();
 180         fiy2             = _mm_setzero_ps();
 181         fiz2             = _mm_setzero_ps();
 182
 183         /* Reset potential sums */
 184         velecsum         = _mm_setzero_ps();
 185         vvdwsum          = _mm_setzero_ps();
 186
 187         /* Start inner kernel loop */
 188         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 189         {
 190
 191             /* Get j neighbor index, and coordinate index */
 192             jnrA             = jjnr[jidx];
 193             jnrB             = jjnr[jidx+1];
 194             jnrC             = jjnr[jidx+2];
 195             jnrD             = jjnr[jidx+3];
 196             j_coord_offsetA  = DIM*jnrA;
 197             j_coord_offsetB  = DIM*jnrB;
 198             j_coord_offsetC  = DIM*jnrC;
 199             j_coord_offsetD  = DIM*jnrD;
 200
 201             /* load j atom coordinates */
 202             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 203                                               x+j_coord_offsetC,x+j_coord_offsetD,
 204                                               &jx0,&jy0,&jz0);
 205
 206             /* Calculate displacement vector */
 207             dx00             = _mm_sub_ps(ix0,jx0);
 208             dy00             = _mm_sub_ps(iy0,jy0);
 209             dz00             = _mm_sub_ps(iz0,jz0);
 210             dx10             = _mm_sub_ps(ix1,jx0);
 211             dy10             = _mm_sub_ps(iy1,jy0);
 212             dz10             = _mm_sub_ps(iz1,jz0);
 213             dx20             = _mm_sub_ps(ix2,jx0);
 214             dy20             = _mm_sub_ps(iy2,jy0);
 215             dz20             = _mm_sub_ps(iz2,jz0);
 216
 217             /* Calculate squared distance and things based on it */
 218             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 219             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 220             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 221
 222             rinv00           = sse2_invsqrt_f(rsq00);
 223             rinv10           = sse2_invsqrt_f(rsq10);
 224             rinv20           = sse2_invsqrt_f(rsq20);
 225
 226             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 227             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 228             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 229
 230             /* Load parameters for j particles */
 231             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 232                                                               charge+jnrC+0,charge+jnrD+0);
 233             vdwjidx0A        = 2*vdwtype[jnrA+0];
 234             vdwjidx0B        = 2*vdwtype[jnrB+0];
 235             vdwjidx0C        = 2*vdwtype[jnrC+0];
 236             vdwjidx0D        = 2*vdwtype[jnrD+0];
 237
 238             fjx0             = _mm_setzero_ps();
 239             fjy0             = _mm_setzero_ps();
 240             fjz0             = _mm_setzero_ps();
 241
 242             /**************************
 243              * CALCULATE INTERACTIONS *
 244              **************************/
 245
 246             if (gmx_mm_any_lt(rsq00,rcutoff2))
 247             {
 248
 249             /* Compute parameters for interactions between i and j atoms */
 250             qq00             = _mm_mul_ps(iq0,jq0);
 251             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 252                                          vdwparam+vdwioffset0+vdwjidx0B,
 253                                          vdwparam+vdwioffset0+vdwjidx0C,
 254                                          vdwparam+vdwioffset0+vdwjidx0D,
 255                                          &c6_00,&c12_00);
 256
 257             /* REACTION-FIELD ELECTROSTATICS */
 258             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 259             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 260
 261             /* LENNARD-JONES DISPERSION/REPULSION */
 262
 263             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 264             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 265             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 266             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 267                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 268             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 269
 270             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 271
 272             /* Update potential sum for this i atom from the interaction with this j atom. */
 273             velec            = _mm_and_ps(velec,cutoff_mask);
 274             velecsum         = _mm_add_ps(velecsum,velec);
 275             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 276             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 277
 278             fscal            = _mm_add_ps(felec,fvdw);
 279
 280             fscal            = _mm_and_ps(fscal,cutoff_mask);
 281
 282             /* Calculate temporary vectorial force */
 283             tx               = _mm_mul_ps(fscal,dx00);
 284             ty               = _mm_mul_ps(fscal,dy00);
 285             tz               = _mm_mul_ps(fscal,dz00);
 286
 287             /* Update vectorial force */
 288             fix0             = _mm_add_ps(fix0,tx);
 289             fiy0             = _mm_add_ps(fiy0,ty);
 290             fiz0             = _mm_add_ps(fiz0,tz);
 291
 292             fjx0             = _mm_add_ps(fjx0,tx);
 293             fjy0             = _mm_add_ps(fjy0,ty);
 294             fjz0             = _mm_add_ps(fjz0,tz);
 295
 296             }
 297
 298             /**************************
 299              * CALCULATE INTERACTIONS *
 300              **************************/
 301
 302             if (gmx_mm_any_lt(rsq10,rcutoff2))
 303             {
 304
 305             /* Compute parameters for interactions between i and j atoms */
 306             qq10             = _mm_mul_ps(iq1,jq0);
 307
 308             /* REACTION-FIELD ELECTROSTATICS */
 309             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 310             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 311
 312             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 313
 314             /* Update potential sum for this i atom from the interaction with this j atom. */
 315             velec            = _mm_and_ps(velec,cutoff_mask);
 316             velecsum         = _mm_add_ps(velecsum,velec);
 317
 318             fscal            = felec;
 319
 320             fscal            = _mm_and_ps(fscal,cutoff_mask);
 321
 322             /* Calculate temporary vectorial force */
 323             tx               = _mm_mul_ps(fscal,dx10);
 324             ty               = _mm_mul_ps(fscal,dy10);
 325             tz               = _mm_mul_ps(fscal,dz10);
 326
 327             /* Update vectorial force */
 328             fix1             = _mm_add_ps(fix1,tx);
 329             fiy1             = _mm_add_ps(fiy1,ty);
 330             fiz1             = _mm_add_ps(fiz1,tz);
 331
 332             fjx0             = _mm_add_ps(fjx0,tx);
 333             fjy0             = _mm_add_ps(fjy0,ty);
 334             fjz0             = _mm_add_ps(fjz0,tz);
 335
 336             }
 337
 338             /**************************
 339              * CALCULATE INTERACTIONS *
 340              **************************/
 341
 342             if (gmx_mm_any_lt(rsq20,rcutoff2))
 343             {
 344
 345             /* Compute parameters for interactions between i and j atoms */
 346             qq20             = _mm_mul_ps(iq2,jq0);
 347
 348             /* REACTION-FIELD ELECTROSTATICS */
 349             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 350             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 351
 352             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 353
 354             /* Update potential sum for this i atom from the interaction with this j atom. */
 355             velec            = _mm_and_ps(velec,cutoff_mask);
 356             velecsum         = _mm_add_ps(velecsum,velec);
 357
 358             fscal            = felec;
 359
 360             fscal            = _mm_and_ps(fscal,cutoff_mask);
 361
 362             /* Calculate temporary vectorial force */
 363             tx               = _mm_mul_ps(fscal,dx20);
 364             ty               = _mm_mul_ps(fscal,dy20);
 365             tz               = _mm_mul_ps(fscal,dz20);
 366
 367             /* Update vectorial force */
 368             fix2             = _mm_add_ps(fix2,tx);
 369             fiy2             = _mm_add_ps(fiy2,ty);
 370             fiz2             = _mm_add_ps(fiz2,tz);
 371
 372             fjx0             = _mm_add_ps(fjx0,tx);
 373             fjy0             = _mm_add_ps(fjy0,ty);
 374             fjz0             = _mm_add_ps(fjz0,tz);
 375
 376             }
 377
 378             fjptrA             = f+j_coord_offsetA;
 379             fjptrB             = f+j_coord_offsetB;
 380             fjptrC             = f+j_coord_offsetC;
 381             fjptrD             = f+j_coord_offsetD;
 382
 383             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 384
 385             /* Inner loop uses 126 flops */
 386         }
 387
 388         if(jidx<j_index_end)
 389         {
 390
 391             /* Get j neighbor index, and coordinate index */
 392             jnrlistA         = jjnr[jidx];
 393             jnrlistB         = jjnr[jidx+1];
 394             jnrlistC         = jjnr[jidx+2];
 395             jnrlistD         = jjnr[jidx+3];
 396             /* Sign of each element will be negative for non-real atoms.
 397              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 398              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 399              */
 400             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 401             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 402             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 403             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 404             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 405             j_coord_offsetA  = DIM*jnrA;
 406             j_coord_offsetB  = DIM*jnrB;
 407             j_coord_offsetC  = DIM*jnrC;
 408             j_coord_offsetD  = DIM*jnrD;
 409
 410             /* load j atom coordinates */
 411             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 412                                               x+j_coord_offsetC,x+j_coord_offsetD,
 413                                               &jx0,&jy0,&jz0);
 414
 415             /* Calculate displacement vector */
 416             dx00             = _mm_sub_ps(ix0,jx0);
 417             dy00             = _mm_sub_ps(iy0,jy0);
 418             dz00             = _mm_sub_ps(iz0,jz0);
 419             dx10             = _mm_sub_ps(ix1,jx0);
 420             dy10             = _mm_sub_ps(iy1,jy0);
 421             dz10             = _mm_sub_ps(iz1,jz0);
 422             dx20             = _mm_sub_ps(ix2,jx0);
 423             dy20             = _mm_sub_ps(iy2,jy0);
 424             dz20             = _mm_sub_ps(iz2,jz0);
 425
 426             /* Calculate squared distance and things based on it */
 427             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 428             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 429             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 430
 431             rinv00           = sse2_invsqrt_f(rsq00);
 432             rinv10           = sse2_invsqrt_f(rsq10);
 433             rinv20           = sse2_invsqrt_f(rsq20);
 434
 435             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 436             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 437             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 438
 439             /* Load parameters for j particles */
 440             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 441                                                               charge+jnrC+0,charge+jnrD+0);
 442             vdwjidx0A        = 2*vdwtype[jnrA+0];
 443             vdwjidx0B        = 2*vdwtype[jnrB+0];
 444             vdwjidx0C        = 2*vdwtype[jnrC+0];
 445             vdwjidx0D        = 2*vdwtype[jnrD+0];
 446
 447             fjx0             = _mm_setzero_ps();
 448             fjy0             = _mm_setzero_ps();
 449             fjz0             = _mm_setzero_ps();
 450
 451             /**************************
 452              * CALCULATE INTERACTIONS *
 453              **************************/
 454
 455             if (gmx_mm_any_lt(rsq00,rcutoff2))
 456             {
 457
 458             /* Compute parameters for interactions between i and j atoms */
 459             qq00             = _mm_mul_ps(iq0,jq0);
 460             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 461                                          vdwparam+vdwioffset0+vdwjidx0B,
 462                                          vdwparam+vdwioffset0+vdwjidx0C,
 463                                          vdwparam+vdwioffset0+vdwjidx0D,
 464                                          &c6_00,&c12_00);
 465
 466             /* REACTION-FIELD ELECTROSTATICS */
 467             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 468             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 469
 470             /* LENNARD-JONES DISPERSION/REPULSION */
 471
 472             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 473             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 474             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 475             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 476                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 477             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 478
 479             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 480
 481             /* Update potential sum for this i atom from the interaction with this j atom. */
 482             velec            = _mm_and_ps(velec,cutoff_mask);
 483             velec            = _mm_andnot_ps(dummy_mask,velec);
 484             velecsum         = _mm_add_ps(velecsum,velec);
 485             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 486             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 487             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 488
 489             fscal            = _mm_add_ps(felec,fvdw);
 490
 491             fscal            = _mm_and_ps(fscal,cutoff_mask);
 492
 493             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 494
 495             /* Calculate temporary vectorial force */
 496             tx               = _mm_mul_ps(fscal,dx00);
 497             ty               = _mm_mul_ps(fscal,dy00);
 498             tz               = _mm_mul_ps(fscal,dz00);
 499
 500             /* Update vectorial force */
 501             fix0             = _mm_add_ps(fix0,tx);
 502             fiy0             = _mm_add_ps(fiy0,ty);
 503             fiz0             = _mm_add_ps(fiz0,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
 511             /**************************
 512              * CALCULATE INTERACTIONS *
 513              **************************/
 514
 515             if (gmx_mm_any_lt(rsq10,rcutoff2))
 516             {
 517
 518             /* Compute parameters for interactions between i and j atoms */
 519             qq10             = _mm_mul_ps(iq1,jq0);
 520
 521             /* REACTION-FIELD ELECTROSTATICS */
 522             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 523             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 524
 525             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 526
 527             /* Update potential sum for this i atom from the interaction with this j atom. */
 528             velec            = _mm_and_ps(velec,cutoff_mask);
 529             velec            = _mm_andnot_ps(dummy_mask,velec);
 530             velecsum         = _mm_add_ps(velecsum,velec);
 531
 532             fscal            = felec;
 533
 534             fscal            = _mm_and_ps(fscal,cutoff_mask);
 535
 536             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 537
 538             /* Calculate temporary vectorial force */
 539             tx               = _mm_mul_ps(fscal,dx10);
 540             ty               = _mm_mul_ps(fscal,dy10);
 541             tz               = _mm_mul_ps(fscal,dz10);
 542
 543             /* Update vectorial force */
 544             fix1             = _mm_add_ps(fix1,tx);
 545             fiy1             = _mm_add_ps(fiy1,ty);
 546             fiz1             = _mm_add_ps(fiz1,tz);
 547
 548             fjx0             = _mm_add_ps(fjx0,tx);
 549             fjy0             = _mm_add_ps(fjy0,ty);
 550             fjz0             = _mm_add_ps(fjz0,tz);
 551
 552             }
 553
 554             /**************************
 555              * CALCULATE INTERACTIONS *
 556              **************************/
 557
 558             if (gmx_mm_any_lt(rsq20,rcutoff2))
 559             {
 560
 561             /* Compute parameters for interactions between i and j atoms */
 562             qq20             = _mm_mul_ps(iq2,jq0);
 563
 564             /* REACTION-FIELD ELECTROSTATICS */
 565             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 566             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 567
 568             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 569
 570             /* Update potential sum for this i atom from the interaction with this j atom. */
 571             velec            = _mm_and_ps(velec,cutoff_mask);
 572             velec            = _mm_andnot_ps(dummy_mask,velec);
 573             velecsum         = _mm_add_ps(velecsum,velec);
 574
 575             fscal            = felec;
 576
 577             fscal            = _mm_and_ps(fscal,cutoff_mask);
 578
 579             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 580
 581             /* Calculate temporary vectorial force */
 582             tx               = _mm_mul_ps(fscal,dx20);
 583             ty               = _mm_mul_ps(fscal,dy20);
 584             tz               = _mm_mul_ps(fscal,dz20);
 585
 586             /* Update vectorial force */
 587             fix2             = _mm_add_ps(fix2,tx);
 588             fiy2             = _mm_add_ps(fiy2,ty);
 589             fiz2             = _mm_add_ps(fiz2,tz);
 590
 591             fjx0             = _mm_add_ps(fjx0,tx);
 592             fjy0             = _mm_add_ps(fjy0,ty);
 593             fjz0             = _mm_add_ps(fjz0,tz);
 594
 595             }
 596
 597             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 598             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 599             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 600             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 601
 602             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 603
 604             /* Inner loop uses 126 flops */
 605         }
 606
 607         /* End of innermost loop */
 608
 609         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 610                                               f+i_coord_offset,fshift+i_shift_offset);
 611
 612         ggid                        = gid[iidx];
 613         /* Update potential energies */
 614         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 615         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 616
 617         /* Increment number of inner iterations */
 618         inneriter                  += j_index_end - j_index_start;
 619
 620         /* Outer loop uses 20 flops */
 621     }
 622
 623     /* Increment number of outer iterations */
 624     outeriter        += nri;
 625
 626     /* Update outer/inner flops */
 627
 628     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*126);
 629 }
 630 /*
 631  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_sse2_single
 632  * Electrostatics interaction: ReactionField
 633  * VdW interaction:            LennardJones
 634  * Geometry:                   Water3-Particle
 635  * Calculate force/pot:        Force
 636  */
 637 void
 638 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_sse2_single
 639                     (t_nblist                    * gmx_restrict       nlist,
 640                      rvec                        * gmx_restrict          xx,
 641                      rvec                        * gmx_restrict          ff,
 642                      struct t_forcerec           * gmx_restrict          fr,
 643                      t_mdatoms                   * gmx_restrict     mdatoms,
 644                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 645                      t_nrnb                      * gmx_restrict        nrnb)
 646 {
 647     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 648      * just 0 for non-waters.
 649      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 650      * jnr indices corresponding to data put in the four positions in the SIMD register.
 651      */
 652     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 653     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 654     int              jnrA,jnrB,jnrC,jnrD;
 655     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 656     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 657     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 658     real             rcutoff_scalar;
 659     real             *shiftvec,*fshift,*x,*f;
 660     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 661     real             scratch[4*DIM];
 662     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 663     int              vdwioffset0;
 664     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 665     int              vdwioffset1;
 666     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 667     int              vdwioffset2;
 668     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 669     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 670     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 671     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 672     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 673     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 674     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 675     real             *charge;
 676     int              nvdwtype;
 677     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 678     int              *vdwtype;
 679     real             *vdwparam;
 680     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 681     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 682     __m128           dummy_mask,cutoff_mask;
 683     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 684     __m128           one     = _mm_set1_ps(1.0);
 685     __m128           two     = _mm_set1_ps(2.0);
 686     x                = xx[0];
 687     f                = ff[0];
 688
 689     nri              = nlist->nri;
 690     iinr             = nlist->iinr;
 691     jindex           = nlist->jindex;
 692     jjnr             = nlist->jjnr;
 693     shiftidx         = nlist->shift;
 694     gid              = nlist->gid;
 695     shiftvec         = fr->shift_vec[0];
 696     fshift           = fr->fshift[0];
 697     facel            = _mm_set1_ps(fr->ic->epsfac);
 698     charge           = mdatoms->chargeA;
 699     krf              = _mm_set1_ps(fr->ic->k_rf);
 700     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 701     crf              = _mm_set1_ps(fr->ic->c_rf);
 702     nvdwtype         = fr->ntype;
 703     vdwparam         = fr->nbfp;
 704     vdwtype          = mdatoms->typeA;
 705
 706     /* Setup water-specific parameters */
 707     inr              = nlist->iinr[0];
 708     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 709     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 710     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 711     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 712
 713     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 714     rcutoff_scalar   = fr->ic->rcoulomb;
 715     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 716     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 717
 718     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 719     rvdw             = _mm_set1_ps(fr->ic->rvdw);
 720
 721     /* Avoid stupid compiler warnings */
 722     jnrA = jnrB = jnrC = jnrD = 0;
 723     j_coord_offsetA = 0;
 724     j_coord_offsetB = 0;
 725     j_coord_offsetC = 0;
 726     j_coord_offsetD = 0;
 727
 728     outeriter        = 0;
 729     inneriter        = 0;
 730
 731     for(iidx=0;iidx<4*DIM;iidx++)
 732     {
 733         scratch[iidx] = 0.0;
 734     }
 735
 736     /* Start outer loop over neighborlists */
 737     for(iidx=0; iidx<nri; iidx++)
 738     {
 739         /* Load shift vector for this list */
 740         i_shift_offset   = DIM*shiftidx[iidx];
 741
 742         /* Load limits for loop over neighbors */
 743         j_index_start    = jindex[iidx];
 744         j_index_end      = jindex[iidx+1];
 745
 746         /* Get outer coordinate index */
 747         inr              = iinr[iidx];
 748         i_coord_offset   = DIM*inr;
 749
 750         /* Load i particle coords and add shift vector */
 751         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 752                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 753
 754         fix0             = _mm_setzero_ps();
 755         fiy0             = _mm_setzero_ps();
 756         fiz0             = _mm_setzero_ps();
 757         fix1             = _mm_setzero_ps();
 758         fiy1             = _mm_setzero_ps();
 759         fiz1             = _mm_setzero_ps();
 760         fix2             = _mm_setzero_ps();
 761         fiy2             = _mm_setzero_ps();
 762         fiz2             = _mm_setzero_ps();
 763
 764         /* Start inner kernel loop */
 765         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 766         {
 767
 768             /* Get j neighbor index, and coordinate index */
 769             jnrA             = jjnr[jidx];
 770             jnrB             = jjnr[jidx+1];
 771             jnrC             = jjnr[jidx+2];
 772             jnrD             = jjnr[jidx+3];
 773             j_coord_offsetA  = DIM*jnrA;
 774             j_coord_offsetB  = DIM*jnrB;
 775             j_coord_offsetC  = DIM*jnrC;
 776             j_coord_offsetD  = DIM*jnrD;
 777
 778             /* load j atom coordinates */
 779             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 780                                               x+j_coord_offsetC,x+j_coord_offsetD,
 781                                               &jx0,&jy0,&jz0);
 782
 783             /* Calculate displacement vector */
 784             dx00             = _mm_sub_ps(ix0,jx0);
 785             dy00             = _mm_sub_ps(iy0,jy0);
 786             dz00             = _mm_sub_ps(iz0,jz0);
 787             dx10             = _mm_sub_ps(ix1,jx0);
 788             dy10             = _mm_sub_ps(iy1,jy0);
 789             dz10             = _mm_sub_ps(iz1,jz0);
 790             dx20             = _mm_sub_ps(ix2,jx0);
 791             dy20             = _mm_sub_ps(iy2,jy0);
 792             dz20             = _mm_sub_ps(iz2,jz0);
 793
 794             /* Calculate squared distance and things based on it */
 795             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 796             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 797             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 798
 799             rinv00           = sse2_invsqrt_f(rsq00);
 800             rinv10           = sse2_invsqrt_f(rsq10);
 801             rinv20           = sse2_invsqrt_f(rsq20);
 802
 803             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 804             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 805             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 806
 807             /* Load parameters for j particles */
 808             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 809                                                               charge+jnrC+0,charge+jnrD+0);
 810             vdwjidx0A        = 2*vdwtype[jnrA+0];
 811             vdwjidx0B        = 2*vdwtype[jnrB+0];
 812             vdwjidx0C        = 2*vdwtype[jnrC+0];
 813             vdwjidx0D        = 2*vdwtype[jnrD+0];
 814
 815             fjx0             = _mm_setzero_ps();
 816             fjy0             = _mm_setzero_ps();
 817             fjz0             = _mm_setzero_ps();
 818
 819             /**************************
 820              * CALCULATE INTERACTIONS *
 821              **************************/
 822
 823             if (gmx_mm_any_lt(rsq00,rcutoff2))
 824             {
 825
 826             /* Compute parameters for interactions between i and j atoms */
 827             qq00             = _mm_mul_ps(iq0,jq0);
 828             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 829                                          vdwparam+vdwioffset0+vdwjidx0B,
 830                                          vdwparam+vdwioffset0+vdwjidx0C,
 831                                          vdwparam+vdwioffset0+vdwjidx0D,
 832                                          &c6_00,&c12_00);
 833
 834             /* REACTION-FIELD ELECTROSTATICS */
 835             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 836
 837             /* LENNARD-JONES DISPERSION/REPULSION */
 838
 839             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 840             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 841
 842             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 843
 844             fscal            = _mm_add_ps(felec,fvdw);
 845
 846             fscal            = _mm_and_ps(fscal,cutoff_mask);
 847
 848             /* Calculate temporary vectorial force */
 849             tx               = _mm_mul_ps(fscal,dx00);
 850             ty               = _mm_mul_ps(fscal,dy00);
 851             tz               = _mm_mul_ps(fscal,dz00);
 852
 853             /* Update vectorial force */
 854             fix0             = _mm_add_ps(fix0,tx);
 855             fiy0             = _mm_add_ps(fiy0,ty);
 856             fiz0             = _mm_add_ps(fiz0,tz);
 857
 858             fjx0             = _mm_add_ps(fjx0,tx);
 859             fjy0             = _mm_add_ps(fjy0,ty);
 860             fjz0             = _mm_add_ps(fjz0,tz);
 861
 862             }
 863
 864             /**************************
 865              * CALCULATE INTERACTIONS *
 866              **************************/
 867
 868             if (gmx_mm_any_lt(rsq10,rcutoff2))
 869             {
 870
 871             /* Compute parameters for interactions between i and j atoms */
 872             qq10             = _mm_mul_ps(iq1,jq0);
 873
 874             /* REACTION-FIELD ELECTROSTATICS */
 875             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 876
 877             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 878
 879             fscal            = felec;
 880
 881             fscal            = _mm_and_ps(fscal,cutoff_mask);
 882
 883             /* Calculate temporary vectorial force */
 884             tx               = _mm_mul_ps(fscal,dx10);
 885             ty               = _mm_mul_ps(fscal,dy10);
 886             tz               = _mm_mul_ps(fscal,dz10);
 887
 888             /* Update vectorial force */
 889             fix1             = _mm_add_ps(fix1,tx);
 890             fiy1             = _mm_add_ps(fiy1,ty);
 891             fiz1             = _mm_add_ps(fiz1,tz);
 892
 893             fjx0             = _mm_add_ps(fjx0,tx);
 894             fjy0             = _mm_add_ps(fjy0,ty);
 895             fjz0             = _mm_add_ps(fjz0,tz);
 896
 897             }
 898
 899             /**************************
 900              * CALCULATE INTERACTIONS *
 901              **************************/
 902
 903             if (gmx_mm_any_lt(rsq20,rcutoff2))
 904             {
 905
 906             /* Compute parameters for interactions between i and j atoms */
 907             qq20             = _mm_mul_ps(iq2,jq0);
 908
 909             /* REACTION-FIELD ELECTROSTATICS */
 910             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 911
 912             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 913
 914             fscal            = felec;
 915
 916             fscal            = _mm_and_ps(fscal,cutoff_mask);
 917
 918             /* Calculate temporary vectorial force */
 919             tx               = _mm_mul_ps(fscal,dx20);
 920             ty               = _mm_mul_ps(fscal,dy20);
 921             tz               = _mm_mul_ps(fscal,dz20);
 922
 923             /* Update vectorial force */
 924             fix2             = _mm_add_ps(fix2,tx);
 925             fiy2             = _mm_add_ps(fiy2,ty);
 926             fiz2             = _mm_add_ps(fiz2,tz);
 927
 928             fjx0             = _mm_add_ps(fjx0,tx);
 929             fjy0             = _mm_add_ps(fjy0,ty);
 930             fjz0             = _mm_add_ps(fjz0,tz);
 931
 932             }
 933
 934             fjptrA             = f+j_coord_offsetA;
 935             fjptrB             = f+j_coord_offsetB;
 936             fjptrC             = f+j_coord_offsetC;
 937             fjptrD             = f+j_coord_offsetD;
 938
 939             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 940
 941             /* Inner loop uses 97 flops */
 942         }
 943
 944         if(jidx<j_index_end)
 945         {
 946
 947             /* Get j neighbor index, and coordinate index */
 948             jnrlistA         = jjnr[jidx];
 949             jnrlistB         = jjnr[jidx+1];
 950             jnrlistC         = jjnr[jidx+2];
 951             jnrlistD         = jjnr[jidx+3];
 952             /* Sign of each element will be negative for non-real atoms.
 953              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 954              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 955              */
 956             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 957             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 958             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 959             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 960             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 961             j_coord_offsetA  = DIM*jnrA;
 962             j_coord_offsetB  = DIM*jnrB;
 963             j_coord_offsetC  = DIM*jnrC;
 964             j_coord_offsetD  = DIM*jnrD;
 965
 966             /* load j atom coordinates */
 967             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 968                                               x+j_coord_offsetC,x+j_coord_offsetD,
 969                                               &jx0,&jy0,&jz0);
 970
 971             /* Calculate displacement vector */
 972             dx00             = _mm_sub_ps(ix0,jx0);
 973             dy00             = _mm_sub_ps(iy0,jy0);
 974             dz00             = _mm_sub_ps(iz0,jz0);
 975             dx10             = _mm_sub_ps(ix1,jx0);
 976             dy10             = _mm_sub_ps(iy1,jy0);
 977             dz10             = _mm_sub_ps(iz1,jz0);
 978             dx20             = _mm_sub_ps(ix2,jx0);
 979             dy20             = _mm_sub_ps(iy2,jy0);
 980             dz20             = _mm_sub_ps(iz2,jz0);
 981
 982             /* Calculate squared distance and things based on it */
 983             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 984             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 985             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 986
 987             rinv00           = sse2_invsqrt_f(rsq00);
 988             rinv10           = sse2_invsqrt_f(rsq10);
 989             rinv20           = sse2_invsqrt_f(rsq20);
 990
 991             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 992             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 993             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 994
 995             /* Load parameters for j particles */
 996             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 997                                                               charge+jnrC+0,charge+jnrD+0);
 998             vdwjidx0A        = 2*vdwtype[jnrA+0];
 999             vdwjidx0B        = 2*vdwtype[jnrB+0];
1000             vdwjidx0C        = 2*vdwtype[jnrC+0];
1001             vdwjidx0D        = 2*vdwtype[jnrD+0];
1002
1003             fjx0             = _mm_setzero_ps();
1004             fjy0             = _mm_setzero_ps();
1005             fjz0             = _mm_setzero_ps();
1006
1007             /**************************
1008              * CALCULATE INTERACTIONS *
1009              **************************/
1010
1011             if (gmx_mm_any_lt(rsq00,rcutoff2))
1012             {
1013
1014             /* Compute parameters for interactions between i and j atoms */
1015             qq00             = _mm_mul_ps(iq0,jq0);
1016             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1017                                          vdwparam+vdwioffset0+vdwjidx0B,
1018                                          vdwparam+vdwioffset0+vdwjidx0C,
1019                                          vdwparam+vdwioffset0+vdwjidx0D,
1020                                          &c6_00,&c12_00);
1021
1022             /* REACTION-FIELD ELECTROSTATICS */
1023             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
1024
1025             /* LENNARD-JONES DISPERSION/REPULSION */
1026
1027             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1028             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1029
1030             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1031
1032             fscal            = _mm_add_ps(felec,fvdw);
1033
1034             fscal            = _mm_and_ps(fscal,cutoff_mask);
1035
1036             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1037
1038             /* Calculate temporary vectorial force */
1039             tx               = _mm_mul_ps(fscal,dx00);
1040             ty               = _mm_mul_ps(fscal,dy00);
1041             tz               = _mm_mul_ps(fscal,dz00);
1042
1043             /* Update vectorial force */
1044             fix0             = _mm_add_ps(fix0,tx);
1045             fiy0             = _mm_add_ps(fiy0,ty);
1046             fiz0             = _mm_add_ps(fiz0,tz);
1047
1048             fjx0             = _mm_add_ps(fjx0,tx);
1049             fjy0             = _mm_add_ps(fjy0,ty);
1050             fjz0             = _mm_add_ps(fjz0,tz);
1051
1052             }
1053
1054             /**************************
1055              * CALCULATE INTERACTIONS *
1056              **************************/
1057
1058             if (gmx_mm_any_lt(rsq10,rcutoff2))
1059             {
1060
1061             /* Compute parameters for interactions between i and j atoms */
1062             qq10             = _mm_mul_ps(iq1,jq0);
1063
1064             /* REACTION-FIELD ELECTROSTATICS */
1065             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1066
1067             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1068
1069             fscal            = felec;
1070
1071             fscal            = _mm_and_ps(fscal,cutoff_mask);
1072
1073             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1074
1075             /* Calculate temporary vectorial force */
1076             tx               = _mm_mul_ps(fscal,dx10);
1077             ty               = _mm_mul_ps(fscal,dy10);
1078             tz               = _mm_mul_ps(fscal,dz10);
1079
1080             /* Update vectorial force */
1081             fix1             = _mm_add_ps(fix1,tx);
1082             fiy1             = _mm_add_ps(fiy1,ty);
1083             fiz1             = _mm_add_ps(fiz1,tz);
1084
1085             fjx0             = _mm_add_ps(fjx0,tx);
1086             fjy0             = _mm_add_ps(fjy0,ty);
1087             fjz0             = _mm_add_ps(fjz0,tz);
1088
1089             }
1090
1091             /**************************
1092              * CALCULATE INTERACTIONS *
1093              **************************/
1094
1095             if (gmx_mm_any_lt(rsq20,rcutoff2))
1096             {
1097
1098             /* Compute parameters for interactions between i and j atoms */
1099             qq20             = _mm_mul_ps(iq2,jq0);
1100
1101             /* REACTION-FIELD ELECTROSTATICS */
1102             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1103
1104             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1105
1106             fscal            = felec;
1107
1108             fscal            = _mm_and_ps(fscal,cutoff_mask);
1109
1110             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1111
1112             /* Calculate temporary vectorial force */
1113             tx               = _mm_mul_ps(fscal,dx20);
1114             ty               = _mm_mul_ps(fscal,dy20);
1115             tz               = _mm_mul_ps(fscal,dz20);
1116
1117             /* Update vectorial force */
1118             fix2             = _mm_add_ps(fix2,tx);
1119             fiy2             = _mm_add_ps(fiy2,ty);
1120             fiz2             = _mm_add_ps(fiz2,tz);
1121
1122             fjx0             = _mm_add_ps(fjx0,tx);
1123             fjy0             = _mm_add_ps(fjy0,ty);
1124             fjz0             = _mm_add_ps(fjz0,tz);
1125
1126             }
1127
1128             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1129             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1130             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1131             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1132
1133             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1134
1135             /* Inner loop uses 97 flops */
1136         }
1137
1138         /* End of innermost loop */
1139
1140         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1141                                               f+i_coord_offset,fshift+i_shift_offset);
1142
1143         /* Increment number of inner iterations */
1144         inneriter                  += j_index_end - j_index_start;
1145
1146         /* Outer loop uses 18 flops */
1147     }
1148
1149     /* Increment number of outer iterations */
1150     outeriter        += nri;
1151
1152     /* Update outer/inner flops */
1153
1154     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*97);
1155 }