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, 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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_single.h"
  48 #include "kernelutil_x86_sse2_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_VF_sse2_single
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_VF_sse2_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  92     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  93     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  94     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 101     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 102     __m128           dummy_mask,cutoff_mask;
 103     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 104     __m128           one     = _mm_set1_ps(1.0);
 105     __m128           two     = _mm_set1_ps(2.0);
 106     x                = xx[0];
 107     f                = ff[0];
 108
 109     nri              = nlist->nri;
 110     iinr             = nlist->iinr;
 111     jindex           = nlist->jindex;
 112     jjnr             = nlist->jjnr;
 113     shiftidx         = nlist->shift;
 114     gid              = nlist->gid;
 115     shiftvec         = fr->shift_vec[0];
 116     fshift           = fr->fshift[0];
 117     facel            = _mm_set1_ps(fr->epsfac);
 118     charge           = mdatoms->chargeA;
 119     krf              = _mm_set1_ps(fr->ic->k_rf);
 120     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 121     crf              = _mm_set1_ps(fr->ic->c_rf);
 122     nvdwtype         = fr->ntype;
 123     vdwparam         = fr->nbfp;
 124     vdwtype          = mdatoms->typeA;
 125
 126     /* Setup water-specific parameters */
 127     inr              = nlist->iinr[0];
 128     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 129     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 130     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 131     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 132
 133     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 134     rcutoff_scalar   = fr->rcoulomb;
 135     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 136     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 137
 138     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 139     rvdw             = _mm_set1_ps(fr->rvdw);
 140
 141     /* Avoid stupid compiler warnings */
 142     jnrA = jnrB = jnrC = jnrD = 0;
 143     j_coord_offsetA = 0;
 144     j_coord_offsetB = 0;
 145     j_coord_offsetC = 0;
 146     j_coord_offsetD = 0;
 147
 148     outeriter        = 0;
 149     inneriter        = 0;
 150
 151     for(iidx=0;iidx<4*DIM;iidx++)
 152     {
 153         scratch[iidx] = 0.0;
 154     }
 155
 156     /* Start outer loop over neighborlists */
 157     for(iidx=0; iidx<nri; iidx++)
 158     {
 159         /* Load shift vector for this list */
 160         i_shift_offset   = DIM*shiftidx[iidx];
 161
 162         /* Load limits for loop over neighbors */
 163         j_index_start    = jindex[iidx];
 164         j_index_end      = jindex[iidx+1];
 165
 166         /* Get outer coordinate index */
 167         inr              = iinr[iidx];
 168         i_coord_offset   = DIM*inr;
 169
 170         /* Load i particle coords and add shift vector */
 171         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 172                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 173
 174         fix0             = _mm_setzero_ps();
 175         fiy0             = _mm_setzero_ps();
 176         fiz0             = _mm_setzero_ps();
 177         fix1             = _mm_setzero_ps();
 178         fiy1             = _mm_setzero_ps();
 179         fiz1             = _mm_setzero_ps();
 180         fix2             = _mm_setzero_ps();
 181         fiy2             = _mm_setzero_ps();
 182         fiz2             = _mm_setzero_ps();
 183
 184         /* Reset potential sums */
 185         velecsum         = _mm_setzero_ps();
 186         vvdwsum          = _mm_setzero_ps();
 187
 188         /* Start inner kernel loop */
 189         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 190         {
 191
 192             /* Get j neighbor index, and coordinate index */
 193             jnrA             = jjnr[jidx];
 194             jnrB             = jjnr[jidx+1];
 195             jnrC             = jjnr[jidx+2];
 196             jnrD             = jjnr[jidx+3];
 197             j_coord_offsetA  = DIM*jnrA;
 198             j_coord_offsetB  = DIM*jnrB;
 199             j_coord_offsetC  = DIM*jnrC;
 200             j_coord_offsetD  = DIM*jnrD;
 201
 202             /* load j atom coordinates */
 203             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 204                                               x+j_coord_offsetC,x+j_coord_offsetD,
 205                                               &jx0,&jy0,&jz0);
 206
 207             /* Calculate displacement vector */
 208             dx00             = _mm_sub_ps(ix0,jx0);
 209             dy00             = _mm_sub_ps(iy0,jy0);
 210             dz00             = _mm_sub_ps(iz0,jz0);
 211             dx10             = _mm_sub_ps(ix1,jx0);
 212             dy10             = _mm_sub_ps(iy1,jy0);
 213             dz10             = _mm_sub_ps(iz1,jz0);
 214             dx20             = _mm_sub_ps(ix2,jx0);
 215             dy20             = _mm_sub_ps(iy2,jy0);
 216             dz20             = _mm_sub_ps(iz2,jz0);
 217
 218             /* Calculate squared distance and things based on it */
 219             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 220             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 221             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 222
 223             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 224             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 225             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 226
 227             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 228             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 229             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 230
 231             /* Load parameters for j particles */
 232             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 233                                                               charge+jnrC+0,charge+jnrD+0);
 234             vdwjidx0A        = 2*vdwtype[jnrA+0];
 235             vdwjidx0B        = 2*vdwtype[jnrB+0];
 236             vdwjidx0C        = 2*vdwtype[jnrC+0];
 237             vdwjidx0D        = 2*vdwtype[jnrD+0];
 238
 239             fjx0             = _mm_setzero_ps();
 240             fjy0             = _mm_setzero_ps();
 241             fjz0             = _mm_setzero_ps();
 242
 243             /**************************
 244              * CALCULATE INTERACTIONS *
 245              **************************/
 246
 247             if (gmx_mm_any_lt(rsq00,rcutoff2))
 248             {
 249
 250             /* Compute parameters for interactions between i and j atoms */
 251             qq00             = _mm_mul_ps(iq0,jq0);
 252             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 253                                          vdwparam+vdwioffset0+vdwjidx0B,
 254                                          vdwparam+vdwioffset0+vdwjidx0C,
 255                                          vdwparam+vdwioffset0+vdwjidx0D,
 256                                          &c6_00,&c12_00);
 257
 258             /* REACTION-FIELD ELECTROSTATICS */
 259             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 260             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 261
 262             /* LENNARD-JONES DISPERSION/REPULSION */
 263
 264             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 265             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 266             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 267             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) ,
 268                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 269             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 270
 271             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 272
 273             /* Update potential sum for this i atom from the interaction with this j atom. */
 274             velec            = _mm_and_ps(velec,cutoff_mask);
 275             velecsum         = _mm_add_ps(velecsum,velec);
 276             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 277             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 278
 279             fscal            = _mm_add_ps(felec,fvdw);
 280
 281             fscal            = _mm_and_ps(fscal,cutoff_mask);
 282
 283             /* Calculate temporary vectorial force */
 284             tx               = _mm_mul_ps(fscal,dx00);
 285             ty               = _mm_mul_ps(fscal,dy00);
 286             tz               = _mm_mul_ps(fscal,dz00);
 287
 288             /* Update vectorial force */
 289             fix0             = _mm_add_ps(fix0,tx);
 290             fiy0             = _mm_add_ps(fiy0,ty);
 291             fiz0             = _mm_add_ps(fiz0,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
 299             /**************************
 300              * CALCULATE INTERACTIONS *
 301              **************************/
 302
 303             if (gmx_mm_any_lt(rsq10,rcutoff2))
 304             {
 305
 306             /* Compute parameters for interactions between i and j atoms */
 307             qq10             = _mm_mul_ps(iq1,jq0);
 308
 309             /* REACTION-FIELD ELECTROSTATICS */
 310             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 311             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 312
 313             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 314
 315             /* Update potential sum for this i atom from the interaction with this j atom. */
 316             velec            = _mm_and_ps(velec,cutoff_mask);
 317             velecsum         = _mm_add_ps(velecsum,velec);
 318
 319             fscal            = felec;
 320
 321             fscal            = _mm_and_ps(fscal,cutoff_mask);
 322
 323             /* Calculate temporary vectorial force */
 324             tx               = _mm_mul_ps(fscal,dx10);
 325             ty               = _mm_mul_ps(fscal,dy10);
 326             tz               = _mm_mul_ps(fscal,dz10);
 327
 328             /* Update vectorial force */
 329             fix1             = _mm_add_ps(fix1,tx);
 330             fiy1             = _mm_add_ps(fiy1,ty);
 331             fiz1             = _mm_add_ps(fiz1,tz);
 332
 333             fjx0             = _mm_add_ps(fjx0,tx);
 334             fjy0             = _mm_add_ps(fjy0,ty);
 335             fjz0             = _mm_add_ps(fjz0,tz);
 336
 337             }
 338
 339             /**************************
 340              * CALCULATE INTERACTIONS *
 341              **************************/
 342
 343             if (gmx_mm_any_lt(rsq20,rcutoff2))
 344             {
 345
 346             /* Compute parameters for interactions between i and j atoms */
 347             qq20             = _mm_mul_ps(iq2,jq0);
 348
 349             /* REACTION-FIELD ELECTROSTATICS */
 350             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 351             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 352
 353             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 354
 355             /* Update potential sum for this i atom from the interaction with this j atom. */
 356             velec            = _mm_and_ps(velec,cutoff_mask);
 357             velecsum         = _mm_add_ps(velecsum,velec);
 358
 359             fscal            = felec;
 360
 361             fscal            = _mm_and_ps(fscal,cutoff_mask);
 362
 363             /* Calculate temporary vectorial force */
 364             tx               = _mm_mul_ps(fscal,dx20);
 365             ty               = _mm_mul_ps(fscal,dy20);
 366             tz               = _mm_mul_ps(fscal,dz20);
 367
 368             /* Update vectorial force */
 369             fix2             = _mm_add_ps(fix2,tx);
 370             fiy2             = _mm_add_ps(fiy2,ty);
 371             fiz2             = _mm_add_ps(fiz2,tz);
 372
 373             fjx0             = _mm_add_ps(fjx0,tx);
 374             fjy0             = _mm_add_ps(fjy0,ty);
 375             fjz0             = _mm_add_ps(fjz0,tz);
 376
 377             }
 378
 379             fjptrA             = f+j_coord_offsetA;
 380             fjptrB             = f+j_coord_offsetB;
 381             fjptrC             = f+j_coord_offsetC;
 382             fjptrD             = f+j_coord_offsetD;
 383
 384             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 385
 386             /* Inner loop uses 126 flops */
 387         }
 388
 389         if(jidx<j_index_end)
 390         {
 391
 392             /* Get j neighbor index, and coordinate index */
 393             jnrlistA         = jjnr[jidx];
 394             jnrlistB         = jjnr[jidx+1];
 395             jnrlistC         = jjnr[jidx+2];
 396             jnrlistD         = jjnr[jidx+3];
 397             /* Sign of each element will be negative for non-real atoms.
 398              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 399              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 400              */
 401             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 402             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 403             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 404             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 405             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 406             j_coord_offsetA  = DIM*jnrA;
 407             j_coord_offsetB  = DIM*jnrB;
 408             j_coord_offsetC  = DIM*jnrC;
 409             j_coord_offsetD  = DIM*jnrD;
 410
 411             /* load j atom coordinates */
 412             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 413                                               x+j_coord_offsetC,x+j_coord_offsetD,
 414                                               &jx0,&jy0,&jz0);
 415
 416             /* Calculate displacement vector */
 417             dx00             = _mm_sub_ps(ix0,jx0);
 418             dy00             = _mm_sub_ps(iy0,jy0);
 419             dz00             = _mm_sub_ps(iz0,jz0);
 420             dx10             = _mm_sub_ps(ix1,jx0);
 421             dy10             = _mm_sub_ps(iy1,jy0);
 422             dz10             = _mm_sub_ps(iz1,jz0);
 423             dx20             = _mm_sub_ps(ix2,jx0);
 424             dy20             = _mm_sub_ps(iy2,jy0);
 425             dz20             = _mm_sub_ps(iz2,jz0);
 426
 427             /* Calculate squared distance and things based on it */
 428             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 429             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 430             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 431
 432             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 433             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 434             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 435
 436             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 437             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 438             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 439
 440             /* Load parameters for j particles */
 441             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 442                                                               charge+jnrC+0,charge+jnrD+0);
 443             vdwjidx0A        = 2*vdwtype[jnrA+0];
 444             vdwjidx0B        = 2*vdwtype[jnrB+0];
 445             vdwjidx0C        = 2*vdwtype[jnrC+0];
 446             vdwjidx0D        = 2*vdwtype[jnrD+0];
 447
 448             fjx0             = _mm_setzero_ps();
 449             fjy0             = _mm_setzero_ps();
 450             fjz0             = _mm_setzero_ps();
 451
 452             /**************************
 453              * CALCULATE INTERACTIONS *
 454              **************************/
 455
 456             if (gmx_mm_any_lt(rsq00,rcutoff2))
 457             {
 458
 459             /* Compute parameters for interactions between i and j atoms */
 460             qq00             = _mm_mul_ps(iq0,jq0);
 461             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 462                                          vdwparam+vdwioffset0+vdwjidx0B,
 463                                          vdwparam+vdwioffset0+vdwjidx0C,
 464                                          vdwparam+vdwioffset0+vdwjidx0D,
 465                                          &c6_00,&c12_00);
 466
 467             /* REACTION-FIELD ELECTROSTATICS */
 468             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 469             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 470
 471             /* LENNARD-JONES DISPERSION/REPULSION */
 472
 473             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 474             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 475             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 476             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) ,
 477                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 478             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 479
 480             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 481
 482             /* Update potential sum for this i atom from the interaction with this j atom. */
 483             velec            = _mm_and_ps(velec,cutoff_mask);
 484             velec            = _mm_andnot_ps(dummy_mask,velec);
 485             velecsum         = _mm_add_ps(velecsum,velec);
 486             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 487             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 488             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 489
 490             fscal            = _mm_add_ps(felec,fvdw);
 491
 492             fscal            = _mm_and_ps(fscal,cutoff_mask);
 493
 494             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 495
 496             /* Calculate temporary vectorial force */
 497             tx               = _mm_mul_ps(fscal,dx00);
 498             ty               = _mm_mul_ps(fscal,dy00);
 499             tz               = _mm_mul_ps(fscal,dz00);
 500
 501             /* Update vectorial force */
 502             fix0             = _mm_add_ps(fix0,tx);
 503             fiy0             = _mm_add_ps(fiy0,ty);
 504             fiz0             = _mm_add_ps(fiz0,tz);
 505
 506             fjx0             = _mm_add_ps(fjx0,tx);
 507             fjy0             = _mm_add_ps(fjy0,ty);
 508             fjz0             = _mm_add_ps(fjz0,tz);
 509
 510             }
 511
 512             /**************************
 513              * CALCULATE INTERACTIONS *
 514              **************************/
 515
 516             if (gmx_mm_any_lt(rsq10,rcutoff2))
 517             {
 518
 519             /* Compute parameters for interactions between i and j atoms */
 520             qq10             = _mm_mul_ps(iq1,jq0);
 521
 522             /* REACTION-FIELD ELECTROSTATICS */
 523             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 524             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 525
 526             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 527
 528             /* Update potential sum for this i atom from the interaction with this j atom. */
 529             velec            = _mm_and_ps(velec,cutoff_mask);
 530             velec            = _mm_andnot_ps(dummy_mask,velec);
 531             velecsum         = _mm_add_ps(velecsum,velec);
 532
 533             fscal            = felec;
 534
 535             fscal            = _mm_and_ps(fscal,cutoff_mask);
 536
 537             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 538
 539             /* Calculate temporary vectorial force */
 540             tx               = _mm_mul_ps(fscal,dx10);
 541             ty               = _mm_mul_ps(fscal,dy10);
 542             tz               = _mm_mul_ps(fscal,dz10);
 543
 544             /* Update vectorial force */
 545             fix1             = _mm_add_ps(fix1,tx);
 546             fiy1             = _mm_add_ps(fiy1,ty);
 547             fiz1             = _mm_add_ps(fiz1,tz);
 548
 549             fjx0             = _mm_add_ps(fjx0,tx);
 550             fjy0             = _mm_add_ps(fjy0,ty);
 551             fjz0             = _mm_add_ps(fjz0,tz);
 552
 553             }
 554
 555             /**************************
 556              * CALCULATE INTERACTIONS *
 557              **************************/
 558
 559             if (gmx_mm_any_lt(rsq20,rcutoff2))
 560             {
 561
 562             /* Compute parameters for interactions between i and j atoms */
 563             qq20             = _mm_mul_ps(iq2,jq0);
 564
 565             /* REACTION-FIELD ELECTROSTATICS */
 566             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 567             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 568
 569             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 570
 571             /* Update potential sum for this i atom from the interaction with this j atom. */
 572             velec            = _mm_and_ps(velec,cutoff_mask);
 573             velec            = _mm_andnot_ps(dummy_mask,velec);
 574             velecsum         = _mm_add_ps(velecsum,velec);
 575
 576             fscal            = felec;
 577
 578             fscal            = _mm_and_ps(fscal,cutoff_mask);
 579
 580             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 581
 582             /* Calculate temporary vectorial force */
 583             tx               = _mm_mul_ps(fscal,dx20);
 584             ty               = _mm_mul_ps(fscal,dy20);
 585             tz               = _mm_mul_ps(fscal,dz20);
 586
 587             /* Update vectorial force */
 588             fix2             = _mm_add_ps(fix2,tx);
 589             fiy2             = _mm_add_ps(fiy2,ty);
 590             fiz2             = _mm_add_ps(fiz2,tz);
 591
 592             fjx0             = _mm_add_ps(fjx0,tx);
 593             fjy0             = _mm_add_ps(fjy0,ty);
 594             fjz0             = _mm_add_ps(fjz0,tz);
 595
 596             }
 597
 598             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 599             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 600             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 601             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 602
 603             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 604
 605             /* Inner loop uses 126 flops */
 606         }
 607
 608         /* End of innermost loop */
 609
 610         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 611                                               f+i_coord_offset,fshift+i_shift_offset);
 612
 613         ggid                        = gid[iidx];
 614         /* Update potential energies */
 615         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 616         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 617
 618         /* Increment number of inner iterations */
 619         inneriter                  += j_index_end - j_index_start;
 620
 621         /* Outer loop uses 20 flops */
 622     }
 623
 624     /* Increment number of outer iterations */
 625     outeriter        += nri;
 626
 627     /* Update outer/inner flops */
 628
 629     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*126);
 630 }
 631 /*
 632  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_sse2_single
 633  * Electrostatics interaction: ReactionField
 634  * VdW interaction:            LennardJones
 635  * Geometry:                   Water3-Particle
 636  * Calculate force/pot:        Force
 637  */
 638 void
 639 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_sse2_single
 640                     (t_nblist                    * gmx_restrict       nlist,
 641                      rvec                        * gmx_restrict          xx,
 642                      rvec                        * gmx_restrict          ff,
 643                      t_forcerec                  * gmx_restrict          fr,
 644                      t_mdatoms                   * gmx_restrict     mdatoms,
 645                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 646                      t_nrnb                      * gmx_restrict        nrnb)
 647 {
 648     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 649      * just 0 for non-waters.
 650      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 651      * jnr indices corresponding to data put in the four positions in the SIMD register.
 652      */
 653     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 654     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 655     int              jnrA,jnrB,jnrC,jnrD;
 656     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 657     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 658     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 659     real             rcutoff_scalar;
 660     real             *shiftvec,*fshift,*x,*f;
 661     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 662     real             scratch[4*DIM];
 663     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 664     int              vdwioffset0;
 665     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 666     int              vdwioffset1;
 667     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 668     int              vdwioffset2;
 669     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 670     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 671     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 672     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 673     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 674     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 675     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 676     real             *charge;
 677     int              nvdwtype;
 678     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 679     int              *vdwtype;
 680     real             *vdwparam;
 681     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 682     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 683     __m128           dummy_mask,cutoff_mask;
 684     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 685     __m128           one     = _mm_set1_ps(1.0);
 686     __m128           two     = _mm_set1_ps(2.0);
 687     x                = xx[0];
 688     f                = ff[0];
 689
 690     nri              = nlist->nri;
 691     iinr             = nlist->iinr;
 692     jindex           = nlist->jindex;
 693     jjnr             = nlist->jjnr;
 694     shiftidx         = nlist->shift;
 695     gid              = nlist->gid;
 696     shiftvec         = fr->shift_vec[0];
 697     fshift           = fr->fshift[0];
 698     facel            = _mm_set1_ps(fr->epsfac);
 699     charge           = mdatoms->chargeA;
 700     krf              = _mm_set1_ps(fr->ic->k_rf);
 701     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 702     crf              = _mm_set1_ps(fr->ic->c_rf);
 703     nvdwtype         = fr->ntype;
 704     vdwparam         = fr->nbfp;
 705     vdwtype          = mdatoms->typeA;
 706
 707     /* Setup water-specific parameters */
 708     inr              = nlist->iinr[0];
 709     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 710     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 711     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 712     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 713
 714     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 715     rcutoff_scalar   = fr->rcoulomb;
 716     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 717     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 718
 719     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 720     rvdw             = _mm_set1_ps(fr->rvdw);
 721
 722     /* Avoid stupid compiler warnings */
 723     jnrA = jnrB = jnrC = jnrD = 0;
 724     j_coord_offsetA = 0;
 725     j_coord_offsetB = 0;
 726     j_coord_offsetC = 0;
 727     j_coord_offsetD = 0;
 728
 729     outeriter        = 0;
 730     inneriter        = 0;
 731
 732     for(iidx=0;iidx<4*DIM;iidx++)
 733     {
 734         scratch[iidx] = 0.0;
 735     }
 736
 737     /* Start outer loop over neighborlists */
 738     for(iidx=0; iidx<nri; iidx++)
 739     {
 740         /* Load shift vector for this list */
 741         i_shift_offset   = DIM*shiftidx[iidx];
 742
 743         /* Load limits for loop over neighbors */
 744         j_index_start    = jindex[iidx];
 745         j_index_end      = jindex[iidx+1];
 746
 747         /* Get outer coordinate index */
 748         inr              = iinr[iidx];
 749         i_coord_offset   = DIM*inr;
 750
 751         /* Load i particle coords and add shift vector */
 752         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 753                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 754
 755         fix0             = _mm_setzero_ps();
 756         fiy0             = _mm_setzero_ps();
 757         fiz0             = _mm_setzero_ps();
 758         fix1             = _mm_setzero_ps();
 759         fiy1             = _mm_setzero_ps();
 760         fiz1             = _mm_setzero_ps();
 761         fix2             = _mm_setzero_ps();
 762         fiy2             = _mm_setzero_ps();
 763         fiz2             = _mm_setzero_ps();
 764
 765         /* Start inner kernel loop */
 766         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 767         {
 768
 769             /* Get j neighbor index, and coordinate index */
 770             jnrA             = jjnr[jidx];
 771             jnrB             = jjnr[jidx+1];
 772             jnrC             = jjnr[jidx+2];
 773             jnrD             = jjnr[jidx+3];
 774             j_coord_offsetA  = DIM*jnrA;
 775             j_coord_offsetB  = DIM*jnrB;
 776             j_coord_offsetC  = DIM*jnrC;
 777             j_coord_offsetD  = DIM*jnrD;
 778
 779             /* load j atom coordinates */
 780             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 781                                               x+j_coord_offsetC,x+j_coord_offsetD,
 782                                               &jx0,&jy0,&jz0);
 783
 784             /* Calculate displacement vector */
 785             dx00             = _mm_sub_ps(ix0,jx0);
 786             dy00             = _mm_sub_ps(iy0,jy0);
 787             dz00             = _mm_sub_ps(iz0,jz0);
 788             dx10             = _mm_sub_ps(ix1,jx0);
 789             dy10             = _mm_sub_ps(iy1,jy0);
 790             dz10             = _mm_sub_ps(iz1,jz0);
 791             dx20             = _mm_sub_ps(ix2,jx0);
 792             dy20             = _mm_sub_ps(iy2,jy0);
 793             dz20             = _mm_sub_ps(iz2,jz0);
 794
 795             /* Calculate squared distance and things based on it */
 796             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 797             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 798             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 799
 800             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 801             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 802             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 803
 804             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 805             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 806             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 807
 808             /* Load parameters for j particles */
 809             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 810                                                               charge+jnrC+0,charge+jnrD+0);
 811             vdwjidx0A        = 2*vdwtype[jnrA+0];
 812             vdwjidx0B        = 2*vdwtype[jnrB+0];
 813             vdwjidx0C        = 2*vdwtype[jnrC+0];
 814             vdwjidx0D        = 2*vdwtype[jnrD+0];
 815
 816             fjx0             = _mm_setzero_ps();
 817             fjy0             = _mm_setzero_ps();
 818             fjz0             = _mm_setzero_ps();
 819
 820             /**************************
 821              * CALCULATE INTERACTIONS *
 822              **************************/
 823
 824             if (gmx_mm_any_lt(rsq00,rcutoff2))
 825             {
 826
 827             /* Compute parameters for interactions between i and j atoms */
 828             qq00             = _mm_mul_ps(iq0,jq0);
 829             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 830                                          vdwparam+vdwioffset0+vdwjidx0B,
 831                                          vdwparam+vdwioffset0+vdwjidx0C,
 832                                          vdwparam+vdwioffset0+vdwjidx0D,
 833                                          &c6_00,&c12_00);
 834
 835             /* REACTION-FIELD ELECTROSTATICS */
 836             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 837
 838             /* LENNARD-JONES DISPERSION/REPULSION */
 839
 840             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 841             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 842
 843             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 844
 845             fscal            = _mm_add_ps(felec,fvdw);
 846
 847             fscal            = _mm_and_ps(fscal,cutoff_mask);
 848
 849             /* Calculate temporary vectorial force */
 850             tx               = _mm_mul_ps(fscal,dx00);
 851             ty               = _mm_mul_ps(fscal,dy00);
 852             tz               = _mm_mul_ps(fscal,dz00);
 853
 854             /* Update vectorial force */
 855             fix0             = _mm_add_ps(fix0,tx);
 856             fiy0             = _mm_add_ps(fiy0,ty);
 857             fiz0             = _mm_add_ps(fiz0,tz);
 858
 859             fjx0             = _mm_add_ps(fjx0,tx);
 860             fjy0             = _mm_add_ps(fjy0,ty);
 861             fjz0             = _mm_add_ps(fjz0,tz);
 862
 863             }
 864
 865             /**************************
 866              * CALCULATE INTERACTIONS *
 867              **************************/
 868
 869             if (gmx_mm_any_lt(rsq10,rcutoff2))
 870             {
 871
 872             /* Compute parameters for interactions between i and j atoms */
 873             qq10             = _mm_mul_ps(iq1,jq0);
 874
 875             /* REACTION-FIELD ELECTROSTATICS */
 876             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 877
 878             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 879
 880             fscal            = felec;
 881
 882             fscal            = _mm_and_ps(fscal,cutoff_mask);
 883
 884             /* Calculate temporary vectorial force */
 885             tx               = _mm_mul_ps(fscal,dx10);
 886             ty               = _mm_mul_ps(fscal,dy10);
 887             tz               = _mm_mul_ps(fscal,dz10);
 888
 889             /* Update vectorial force */
 890             fix1             = _mm_add_ps(fix1,tx);
 891             fiy1             = _mm_add_ps(fiy1,ty);
 892             fiz1             = _mm_add_ps(fiz1,tz);
 893
 894             fjx0             = _mm_add_ps(fjx0,tx);
 895             fjy0             = _mm_add_ps(fjy0,ty);
 896             fjz0             = _mm_add_ps(fjz0,tz);
 897
 898             }
 899
 900             /**************************
 901              * CALCULATE INTERACTIONS *
 902              **************************/
 903
 904             if (gmx_mm_any_lt(rsq20,rcutoff2))
 905             {
 906
 907             /* Compute parameters for interactions between i and j atoms */
 908             qq20             = _mm_mul_ps(iq2,jq0);
 909
 910             /* REACTION-FIELD ELECTROSTATICS */
 911             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 912
 913             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 914
 915             fscal            = felec;
 916
 917             fscal            = _mm_and_ps(fscal,cutoff_mask);
 918
 919             /* Calculate temporary vectorial force */
 920             tx               = _mm_mul_ps(fscal,dx20);
 921             ty               = _mm_mul_ps(fscal,dy20);
 922             tz               = _mm_mul_ps(fscal,dz20);
 923
 924             /* Update vectorial force */
 925             fix2             = _mm_add_ps(fix2,tx);
 926             fiy2             = _mm_add_ps(fiy2,ty);
 927             fiz2             = _mm_add_ps(fiz2,tz);
 928
 929             fjx0             = _mm_add_ps(fjx0,tx);
 930             fjy0             = _mm_add_ps(fjy0,ty);
 931             fjz0             = _mm_add_ps(fjz0,tz);
 932
 933             }
 934
 935             fjptrA             = f+j_coord_offsetA;
 936             fjptrB             = f+j_coord_offsetB;
 937             fjptrC             = f+j_coord_offsetC;
 938             fjptrD             = f+j_coord_offsetD;
 939
 940             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 941
 942             /* Inner loop uses 97 flops */
 943         }
 944
 945         if(jidx<j_index_end)
 946         {
 947
 948             /* Get j neighbor index, and coordinate index */
 949             jnrlistA         = jjnr[jidx];
 950             jnrlistB         = jjnr[jidx+1];
 951             jnrlistC         = jjnr[jidx+2];
 952             jnrlistD         = jjnr[jidx+3];
 953             /* Sign of each element will be negative for non-real atoms.
 954              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 955              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 956              */
 957             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 958             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 959             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 960             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 961             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 962             j_coord_offsetA  = DIM*jnrA;
 963             j_coord_offsetB  = DIM*jnrB;
 964             j_coord_offsetC  = DIM*jnrC;
 965             j_coord_offsetD  = DIM*jnrD;
 966
 967             /* load j atom coordinates */
 968             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 969                                               x+j_coord_offsetC,x+j_coord_offsetD,
 970                                               &jx0,&jy0,&jz0);
 971
 972             /* Calculate displacement vector */
 973             dx00             = _mm_sub_ps(ix0,jx0);
 974             dy00             = _mm_sub_ps(iy0,jy0);
 975             dz00             = _mm_sub_ps(iz0,jz0);
 976             dx10             = _mm_sub_ps(ix1,jx0);
 977             dy10             = _mm_sub_ps(iy1,jy0);
 978             dz10             = _mm_sub_ps(iz1,jz0);
 979             dx20             = _mm_sub_ps(ix2,jx0);
 980             dy20             = _mm_sub_ps(iy2,jy0);
 981             dz20             = _mm_sub_ps(iz2,jz0);
 982
 983             /* Calculate squared distance and things based on it */
 984             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 985             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 986             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 987
 988             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 989             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 990             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 991
 992             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 993             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 994             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 995
 996             /* Load parameters for j particles */
 997             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 998                                                               charge+jnrC+0,charge+jnrD+0);
 999             vdwjidx0A        = 2*vdwtype[jnrA+0];
1000             vdwjidx0B        = 2*vdwtype[jnrB+0];
1001             vdwjidx0C        = 2*vdwtype[jnrC+0];
1002             vdwjidx0D        = 2*vdwtype[jnrD+0];
1003
1004             fjx0             = _mm_setzero_ps();
1005             fjy0             = _mm_setzero_ps();
1006             fjz0             = _mm_setzero_ps();
1007
1008             /**************************
1009              * CALCULATE INTERACTIONS *
1010              **************************/
1011
1012             if (gmx_mm_any_lt(rsq00,rcutoff2))
1013             {
1014
1015             /* Compute parameters for interactions between i and j atoms */
1016             qq00             = _mm_mul_ps(iq0,jq0);
1017             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1018                                          vdwparam+vdwioffset0+vdwjidx0B,
1019                                          vdwparam+vdwioffset0+vdwjidx0C,
1020                                          vdwparam+vdwioffset0+vdwjidx0D,
1021                                          &c6_00,&c12_00);
1022
1023             /* REACTION-FIELD ELECTROSTATICS */
1024             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
1025
1026             /* LENNARD-JONES DISPERSION/REPULSION */
1027
1028             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1029             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1030
1031             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1032
1033             fscal            = _mm_add_ps(felec,fvdw);
1034
1035             fscal            = _mm_and_ps(fscal,cutoff_mask);
1036
1037             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1038
1039             /* Calculate temporary vectorial force */
1040             tx               = _mm_mul_ps(fscal,dx00);
1041             ty               = _mm_mul_ps(fscal,dy00);
1042             tz               = _mm_mul_ps(fscal,dz00);
1043
1044             /* Update vectorial force */
1045             fix0             = _mm_add_ps(fix0,tx);
1046             fiy0             = _mm_add_ps(fiy0,ty);
1047             fiz0             = _mm_add_ps(fiz0,tz);
1048
1049             fjx0             = _mm_add_ps(fjx0,tx);
1050             fjy0             = _mm_add_ps(fjy0,ty);
1051             fjz0             = _mm_add_ps(fjz0,tz);
1052
1053             }
1054
1055             /**************************
1056              * CALCULATE INTERACTIONS *
1057              **************************/
1058
1059             if (gmx_mm_any_lt(rsq10,rcutoff2))
1060             {
1061
1062             /* Compute parameters for interactions between i and j atoms */
1063             qq10             = _mm_mul_ps(iq1,jq0);
1064
1065             /* REACTION-FIELD ELECTROSTATICS */
1066             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1067
1068             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1069
1070             fscal            = felec;
1071
1072             fscal            = _mm_and_ps(fscal,cutoff_mask);
1073
1074             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1075
1076             /* Calculate temporary vectorial force */
1077             tx               = _mm_mul_ps(fscal,dx10);
1078             ty               = _mm_mul_ps(fscal,dy10);
1079             tz               = _mm_mul_ps(fscal,dz10);
1080
1081             /* Update vectorial force */
1082             fix1             = _mm_add_ps(fix1,tx);
1083             fiy1             = _mm_add_ps(fiy1,ty);
1084             fiz1             = _mm_add_ps(fiz1,tz);
1085
1086             fjx0             = _mm_add_ps(fjx0,tx);
1087             fjy0             = _mm_add_ps(fjy0,ty);
1088             fjz0             = _mm_add_ps(fjz0,tz);
1089
1090             }
1091
1092             /**************************
1093              * CALCULATE INTERACTIONS *
1094              **************************/
1095
1096             if (gmx_mm_any_lt(rsq20,rcutoff2))
1097             {
1098
1099             /* Compute parameters for interactions between i and j atoms */
1100             qq20             = _mm_mul_ps(iq2,jq0);
1101
1102             /* REACTION-FIELD ELECTROSTATICS */
1103             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1104
1105             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1106
1107             fscal            = felec;
1108
1109             fscal            = _mm_and_ps(fscal,cutoff_mask);
1110
1111             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1112
1113             /* Calculate temporary vectorial force */
1114             tx               = _mm_mul_ps(fscal,dx20);
1115             ty               = _mm_mul_ps(fscal,dy20);
1116             tz               = _mm_mul_ps(fscal,dz20);
1117
1118             /* Update vectorial force */
1119             fix2             = _mm_add_ps(fix2,tx);
1120             fiy2             = _mm_add_ps(fiy2,ty);
1121             fiz2             = _mm_add_ps(fiz2,tz);
1122
1123             fjx0             = _mm_add_ps(fjx0,tx);
1124             fjy0             = _mm_add_ps(fjy0,ty);
1125             fjz0             = _mm_add_ps(fjz0,tz);
1126
1127             }
1128
1129             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1130             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1131             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1132             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1133
1134             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1135
1136             /* Inner loop uses 97 flops */
1137         }
1138
1139         /* End of innermost loop */
1140
1141         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1142                                               f+i_coord_offset,fshift+i_shift_offset);
1143
1144         /* Increment number of inner iterations */
1145         inneriter                  += j_index_end - j_index_start;
1146
1147         /* Outer loop uses 18 flops */
1148     }
1149
1150     /* Increment number of outer iterations */
1151     outeriter        += nri;
1152
1153     /* Update outer/inner flops */
1154
1155     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*97);
1156 }