src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRFCut_VdwLJSw_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_VdwLJSw_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_VdwLJSw_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           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 103     real             rswitch_scalar,d_scalar;
 104     __m128           dummy_mask,cutoff_mask;
 105     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 106     __m128           one     = _mm_set1_ps(1.0);
 107     __m128           two     = _mm_set1_ps(2.0);
 108     x                = xx[0];
 109     f                = ff[0];
 110
 111     nri              = nlist->nri;
 112     iinr             = nlist->iinr;
 113     jindex           = nlist->jindex;
 114     jjnr             = nlist->jjnr;
 115     shiftidx         = nlist->shift;
 116     gid              = nlist->gid;
 117     shiftvec         = fr->shift_vec[0];
 118     fshift           = fr->fshift[0];
 119     facel            = _mm_set1_ps(fr->epsfac);
 120     charge           = mdatoms->chargeA;
 121     krf              = _mm_set1_ps(fr->ic->k_rf);
 122     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 123     crf              = _mm_set1_ps(fr->ic->c_rf);
 124     nvdwtype         = fr->ntype;
 125     vdwparam         = fr->nbfp;
 126     vdwtype          = mdatoms->typeA;
 127
 128     /* Setup water-specific parameters */
 129     inr              = nlist->iinr[0];
 130     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 131     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 132     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 133     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 134
 135     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 136     rcutoff_scalar   = fr->rcoulomb;
 137     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 138     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 139
 140     rswitch_scalar   = fr->rvdw_switch;
 141     rswitch          = _mm_set1_ps(rswitch_scalar);
 142     /* Setup switch parameters */
 143     d_scalar         = rcutoff_scalar-rswitch_scalar;
 144     d                = _mm_set1_ps(d_scalar);
 145     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 146     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 147     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 148     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 149     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 150     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 151
 152     /* Avoid stupid compiler warnings */
 153     jnrA = jnrB = jnrC = jnrD = 0;
 154     j_coord_offsetA = 0;
 155     j_coord_offsetB = 0;
 156     j_coord_offsetC = 0;
 157     j_coord_offsetD = 0;
 158
 159     outeriter        = 0;
 160     inneriter        = 0;
 161
 162     for(iidx=0;iidx<4*DIM;iidx++)
 163     {
 164         scratch[iidx] = 0.0;
 165     }
 166
 167     /* Start outer loop over neighborlists */
 168     for(iidx=0; iidx<nri; iidx++)
 169     {
 170         /* Load shift vector for this list */
 171         i_shift_offset   = DIM*shiftidx[iidx];
 172
 173         /* Load limits for loop over neighbors */
 174         j_index_start    = jindex[iidx];
 175         j_index_end      = jindex[iidx+1];
 176
 177         /* Get outer coordinate index */
 178         inr              = iinr[iidx];
 179         i_coord_offset   = DIM*inr;
 180
 181         /* Load i particle coords and add shift vector */
 182         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 183                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 184
 185         fix0             = _mm_setzero_ps();
 186         fiy0             = _mm_setzero_ps();
 187         fiz0             = _mm_setzero_ps();
 188         fix1             = _mm_setzero_ps();
 189         fiy1             = _mm_setzero_ps();
 190         fiz1             = _mm_setzero_ps();
 191         fix2             = _mm_setzero_ps();
 192         fiy2             = _mm_setzero_ps();
 193         fiz2             = _mm_setzero_ps();
 194
 195         /* Reset potential sums */
 196         velecsum         = _mm_setzero_ps();
 197         vvdwsum          = _mm_setzero_ps();
 198
 199         /* Start inner kernel loop */
 200         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 201         {
 202
 203             /* Get j neighbor index, and coordinate index */
 204             jnrA             = jjnr[jidx];
 205             jnrB             = jjnr[jidx+1];
 206             jnrC             = jjnr[jidx+2];
 207             jnrD             = jjnr[jidx+3];
 208             j_coord_offsetA  = DIM*jnrA;
 209             j_coord_offsetB  = DIM*jnrB;
 210             j_coord_offsetC  = DIM*jnrC;
 211             j_coord_offsetD  = DIM*jnrD;
 212
 213             /* load j atom coordinates */
 214             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 215                                               x+j_coord_offsetC,x+j_coord_offsetD,
 216                                               &jx0,&jy0,&jz0);
 217
 218             /* Calculate displacement vector */
 219             dx00             = _mm_sub_ps(ix0,jx0);
 220             dy00             = _mm_sub_ps(iy0,jy0);
 221             dz00             = _mm_sub_ps(iz0,jz0);
 222             dx10             = _mm_sub_ps(ix1,jx0);
 223             dy10             = _mm_sub_ps(iy1,jy0);
 224             dz10             = _mm_sub_ps(iz1,jz0);
 225             dx20             = _mm_sub_ps(ix2,jx0);
 226             dy20             = _mm_sub_ps(iy2,jy0);
 227             dz20             = _mm_sub_ps(iz2,jz0);
 228
 229             /* Calculate squared distance and things based on it */
 230             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 231             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 232             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 233
 234             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 235             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 236             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 237
 238             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 239             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 240             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 241
 242             /* Load parameters for j particles */
 243             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 244                                                               charge+jnrC+0,charge+jnrD+0);
 245             vdwjidx0A        = 2*vdwtype[jnrA+0];
 246             vdwjidx0B        = 2*vdwtype[jnrB+0];
 247             vdwjidx0C        = 2*vdwtype[jnrC+0];
 248             vdwjidx0D        = 2*vdwtype[jnrD+0];
 249
 250             fjx0             = _mm_setzero_ps();
 251             fjy0             = _mm_setzero_ps();
 252             fjz0             = _mm_setzero_ps();
 253
 254             /**************************
 255              * CALCULATE INTERACTIONS *
 256              **************************/
 257
 258             if (gmx_mm_any_lt(rsq00,rcutoff2))
 259             {
 260
 261             r00              = _mm_mul_ps(rsq00,rinv00);
 262
 263             /* Compute parameters for interactions between i and j atoms */
 264             qq00             = _mm_mul_ps(iq0,jq0);
 265             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 266                                          vdwparam+vdwioffset0+vdwjidx0B,
 267                                          vdwparam+vdwioffset0+vdwjidx0C,
 268                                          vdwparam+vdwioffset0+vdwjidx0D,
 269                                          &c6_00,&c12_00);
 270
 271             /* REACTION-FIELD ELECTROSTATICS */
 272             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 273             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 274
 275             /* LENNARD-JONES DISPERSION/REPULSION */
 276
 277             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 278             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 279             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 280             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 281             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 282
 283             d                = _mm_sub_ps(r00,rswitch);
 284             d                = _mm_max_ps(d,_mm_setzero_ps());
 285             d2               = _mm_mul_ps(d,d);
 286             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 287
 288             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 289
 290             /* Evaluate switch function */
 291             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 292             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 293             vvdw             = _mm_mul_ps(vvdw,sw);
 294             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 295
 296             /* Update potential sum for this i atom from the interaction with this j atom. */
 297             velec            = _mm_and_ps(velec,cutoff_mask);
 298             velecsum         = _mm_add_ps(velecsum,velec);
 299             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 300             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 301
 302             fscal            = _mm_add_ps(felec,fvdw);
 303
 304             fscal            = _mm_and_ps(fscal,cutoff_mask);
 305
 306             /* Calculate temporary vectorial force */
 307             tx               = _mm_mul_ps(fscal,dx00);
 308             ty               = _mm_mul_ps(fscal,dy00);
 309             tz               = _mm_mul_ps(fscal,dz00);
 310
 311             /* Update vectorial force */
 312             fix0             = _mm_add_ps(fix0,tx);
 313             fiy0             = _mm_add_ps(fiy0,ty);
 314             fiz0             = _mm_add_ps(fiz0,tz);
 315
 316             fjx0             = _mm_add_ps(fjx0,tx);
 317             fjy0             = _mm_add_ps(fjy0,ty);
 318             fjz0             = _mm_add_ps(fjz0,tz);
 319
 320             }
 321
 322             /**************************
 323              * CALCULATE INTERACTIONS *
 324              **************************/
 325
 326             if (gmx_mm_any_lt(rsq10,rcutoff2))
 327             {
 328
 329             /* Compute parameters for interactions between i and j atoms */
 330             qq10             = _mm_mul_ps(iq1,jq0);
 331
 332             /* REACTION-FIELD ELECTROSTATICS */
 333             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 334             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 335
 336             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 337
 338             /* Update potential sum for this i atom from the interaction with this j atom. */
 339             velec            = _mm_and_ps(velec,cutoff_mask);
 340             velecsum         = _mm_add_ps(velecsum,velec);
 341
 342             fscal            = felec;
 343
 344             fscal            = _mm_and_ps(fscal,cutoff_mask);
 345
 346             /* Calculate temporary vectorial force */
 347             tx               = _mm_mul_ps(fscal,dx10);
 348             ty               = _mm_mul_ps(fscal,dy10);
 349             tz               = _mm_mul_ps(fscal,dz10);
 350
 351             /* Update vectorial force */
 352             fix1             = _mm_add_ps(fix1,tx);
 353             fiy1             = _mm_add_ps(fiy1,ty);
 354             fiz1             = _mm_add_ps(fiz1,tz);
 355
 356             fjx0             = _mm_add_ps(fjx0,tx);
 357             fjy0             = _mm_add_ps(fjy0,ty);
 358             fjz0             = _mm_add_ps(fjz0,tz);
 359
 360             }
 361
 362             /**************************
 363              * CALCULATE INTERACTIONS *
 364              **************************/
 365
 366             if (gmx_mm_any_lt(rsq20,rcutoff2))
 367             {
 368
 369             /* Compute parameters for interactions between i and j atoms */
 370             qq20             = _mm_mul_ps(iq2,jq0);
 371
 372             /* REACTION-FIELD ELECTROSTATICS */
 373             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 374             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 375
 376             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 377
 378             /* Update potential sum for this i atom from the interaction with this j atom. */
 379             velec            = _mm_and_ps(velec,cutoff_mask);
 380             velecsum         = _mm_add_ps(velecsum,velec);
 381
 382             fscal            = felec;
 383
 384             fscal            = _mm_and_ps(fscal,cutoff_mask);
 385
 386             /* Calculate temporary vectorial force */
 387             tx               = _mm_mul_ps(fscal,dx20);
 388             ty               = _mm_mul_ps(fscal,dy20);
 389             tz               = _mm_mul_ps(fscal,dz20);
 390
 391             /* Update vectorial force */
 392             fix2             = _mm_add_ps(fix2,tx);
 393             fiy2             = _mm_add_ps(fiy2,ty);
 394             fiz2             = _mm_add_ps(fiz2,tz);
 395
 396             fjx0             = _mm_add_ps(fjx0,tx);
 397             fjy0             = _mm_add_ps(fjy0,ty);
 398             fjz0             = _mm_add_ps(fjz0,tz);
 399
 400             }
 401
 402             fjptrA             = f+j_coord_offsetA;
 403             fjptrB             = f+j_coord_offsetB;
 404             fjptrC             = f+j_coord_offsetC;
 405             fjptrD             = f+j_coord_offsetD;
 406
 407             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 408
 409             /* Inner loop uses 142 flops */
 410         }
 411
 412         if(jidx<j_index_end)
 413         {
 414
 415             /* Get j neighbor index, and coordinate index */
 416             jnrlistA         = jjnr[jidx];
 417             jnrlistB         = jjnr[jidx+1];
 418             jnrlistC         = jjnr[jidx+2];
 419             jnrlistD         = jjnr[jidx+3];
 420             /* Sign of each element will be negative for non-real atoms.
 421              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 422              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 423              */
 424             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 425             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 426             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 427             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 428             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 429             j_coord_offsetA  = DIM*jnrA;
 430             j_coord_offsetB  = DIM*jnrB;
 431             j_coord_offsetC  = DIM*jnrC;
 432             j_coord_offsetD  = DIM*jnrD;
 433
 434             /* load j atom coordinates */
 435             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 436                                               x+j_coord_offsetC,x+j_coord_offsetD,
 437                                               &jx0,&jy0,&jz0);
 438
 439             /* Calculate displacement vector */
 440             dx00             = _mm_sub_ps(ix0,jx0);
 441             dy00             = _mm_sub_ps(iy0,jy0);
 442             dz00             = _mm_sub_ps(iz0,jz0);
 443             dx10             = _mm_sub_ps(ix1,jx0);
 444             dy10             = _mm_sub_ps(iy1,jy0);
 445             dz10             = _mm_sub_ps(iz1,jz0);
 446             dx20             = _mm_sub_ps(ix2,jx0);
 447             dy20             = _mm_sub_ps(iy2,jy0);
 448             dz20             = _mm_sub_ps(iz2,jz0);
 449
 450             /* Calculate squared distance and things based on it */
 451             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 452             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 453             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 454
 455             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 456             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 457             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 458
 459             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 460             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 461             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 462
 463             /* Load parameters for j particles */
 464             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 465                                                               charge+jnrC+0,charge+jnrD+0);
 466             vdwjidx0A        = 2*vdwtype[jnrA+0];
 467             vdwjidx0B        = 2*vdwtype[jnrB+0];
 468             vdwjidx0C        = 2*vdwtype[jnrC+0];
 469             vdwjidx0D        = 2*vdwtype[jnrD+0];
 470
 471             fjx0             = _mm_setzero_ps();
 472             fjy0             = _mm_setzero_ps();
 473             fjz0             = _mm_setzero_ps();
 474
 475             /**************************
 476              * CALCULATE INTERACTIONS *
 477              **************************/
 478
 479             if (gmx_mm_any_lt(rsq00,rcutoff2))
 480             {
 481
 482             r00              = _mm_mul_ps(rsq00,rinv00);
 483             r00              = _mm_andnot_ps(dummy_mask,r00);
 484
 485             /* Compute parameters for interactions between i and j atoms */
 486             qq00             = _mm_mul_ps(iq0,jq0);
 487             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 488                                          vdwparam+vdwioffset0+vdwjidx0B,
 489                                          vdwparam+vdwioffset0+vdwjidx0C,
 490                                          vdwparam+vdwioffset0+vdwjidx0D,
 491                                          &c6_00,&c12_00);
 492
 493             /* REACTION-FIELD ELECTROSTATICS */
 494             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 495             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 496
 497             /* LENNARD-JONES DISPERSION/REPULSION */
 498
 499             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 500             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 501             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 502             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 503             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 504
 505             d                = _mm_sub_ps(r00,rswitch);
 506             d                = _mm_max_ps(d,_mm_setzero_ps());
 507             d2               = _mm_mul_ps(d,d);
 508             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 509
 510             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 511
 512             /* Evaluate switch function */
 513             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 514             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 515             vvdw             = _mm_mul_ps(vvdw,sw);
 516             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 517
 518             /* Update potential sum for this i atom from the interaction with this j atom. */
 519             velec            = _mm_and_ps(velec,cutoff_mask);
 520             velec            = _mm_andnot_ps(dummy_mask,velec);
 521             velecsum         = _mm_add_ps(velecsum,velec);
 522             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 523             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 524             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 525
 526             fscal            = _mm_add_ps(felec,fvdw);
 527
 528             fscal            = _mm_and_ps(fscal,cutoff_mask);
 529
 530             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 531
 532             /* Calculate temporary vectorial force */
 533             tx               = _mm_mul_ps(fscal,dx00);
 534             ty               = _mm_mul_ps(fscal,dy00);
 535             tz               = _mm_mul_ps(fscal,dz00);
 536
 537             /* Update vectorial force */
 538             fix0             = _mm_add_ps(fix0,tx);
 539             fiy0             = _mm_add_ps(fiy0,ty);
 540             fiz0             = _mm_add_ps(fiz0,tz);
 541
 542             fjx0             = _mm_add_ps(fjx0,tx);
 543             fjy0             = _mm_add_ps(fjy0,ty);
 544             fjz0             = _mm_add_ps(fjz0,tz);
 545
 546             }
 547
 548             /**************************
 549              * CALCULATE INTERACTIONS *
 550              **************************/
 551
 552             if (gmx_mm_any_lt(rsq10,rcutoff2))
 553             {
 554
 555             /* Compute parameters for interactions between i and j atoms */
 556             qq10             = _mm_mul_ps(iq1,jq0);
 557
 558             /* REACTION-FIELD ELECTROSTATICS */
 559             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 560             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 561
 562             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 563
 564             /* Update potential sum for this i atom from the interaction with this j atom. */
 565             velec            = _mm_and_ps(velec,cutoff_mask);
 566             velec            = _mm_andnot_ps(dummy_mask,velec);
 567             velecsum         = _mm_add_ps(velecsum,velec);
 568
 569             fscal            = felec;
 570
 571             fscal            = _mm_and_ps(fscal,cutoff_mask);
 572
 573             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 574
 575             /* Calculate temporary vectorial force */
 576             tx               = _mm_mul_ps(fscal,dx10);
 577             ty               = _mm_mul_ps(fscal,dy10);
 578             tz               = _mm_mul_ps(fscal,dz10);
 579
 580             /* Update vectorial force */
 581             fix1             = _mm_add_ps(fix1,tx);
 582             fiy1             = _mm_add_ps(fiy1,ty);
 583             fiz1             = _mm_add_ps(fiz1,tz);
 584
 585             fjx0             = _mm_add_ps(fjx0,tx);
 586             fjy0             = _mm_add_ps(fjy0,ty);
 587             fjz0             = _mm_add_ps(fjz0,tz);
 588
 589             }
 590
 591             /**************************
 592              * CALCULATE INTERACTIONS *
 593              **************************/
 594
 595             if (gmx_mm_any_lt(rsq20,rcutoff2))
 596             {
 597
 598             /* Compute parameters for interactions between i and j atoms */
 599             qq20             = _mm_mul_ps(iq2,jq0);
 600
 601             /* REACTION-FIELD ELECTROSTATICS */
 602             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 603             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 604
 605             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 606
 607             /* Update potential sum for this i atom from the interaction with this j atom. */
 608             velec            = _mm_and_ps(velec,cutoff_mask);
 609             velec            = _mm_andnot_ps(dummy_mask,velec);
 610             velecsum         = _mm_add_ps(velecsum,velec);
 611
 612             fscal            = felec;
 613
 614             fscal            = _mm_and_ps(fscal,cutoff_mask);
 615
 616             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 617
 618             /* Calculate temporary vectorial force */
 619             tx               = _mm_mul_ps(fscal,dx20);
 620             ty               = _mm_mul_ps(fscal,dy20);
 621             tz               = _mm_mul_ps(fscal,dz20);
 622
 623             /* Update vectorial force */
 624             fix2             = _mm_add_ps(fix2,tx);
 625             fiy2             = _mm_add_ps(fiy2,ty);
 626             fiz2             = _mm_add_ps(fiz2,tz);
 627
 628             fjx0             = _mm_add_ps(fjx0,tx);
 629             fjy0             = _mm_add_ps(fjy0,ty);
 630             fjz0             = _mm_add_ps(fjz0,tz);
 631
 632             }
 633
 634             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 635             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 636             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 637             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 638
 639             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 640
 641             /* Inner loop uses 143 flops */
 642         }
 643
 644         /* End of innermost loop */
 645
 646         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 647                                               f+i_coord_offset,fshift+i_shift_offset);
 648
 649         ggid                        = gid[iidx];
 650         /* Update potential energies */
 651         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 652         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 653
 654         /* Increment number of inner iterations */
 655         inneriter                  += j_index_end - j_index_start;
 656
 657         /* Outer loop uses 20 flops */
 658     }
 659
 660     /* Increment number of outer iterations */
 661     outeriter        += nri;
 662
 663     /* Update outer/inner flops */
 664
 665     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*143);
 666 }
 667 /*
 668  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_sse2_single
 669  * Electrostatics interaction: ReactionField
 670  * VdW interaction:            LennardJones
 671  * Geometry:                   Water3-Particle
 672  * Calculate force/pot:        Force
 673  */
 674 void
 675 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_sse2_single
 676                     (t_nblist                    * gmx_restrict       nlist,
 677                      rvec                        * gmx_restrict          xx,
 678                      rvec                        * gmx_restrict          ff,
 679                      t_forcerec                  * gmx_restrict          fr,
 680                      t_mdatoms                   * gmx_restrict     mdatoms,
 681                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 682                      t_nrnb                      * gmx_restrict        nrnb)
 683 {
 684     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 685      * just 0 for non-waters.
 686      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 687      * jnr indices corresponding to data put in the four positions in the SIMD register.
 688      */
 689     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 690     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 691     int              jnrA,jnrB,jnrC,jnrD;
 692     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 693     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 694     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 695     real             rcutoff_scalar;
 696     real             *shiftvec,*fshift,*x,*f;
 697     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 698     real             scratch[4*DIM];
 699     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 700     int              vdwioffset0;
 701     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 702     int              vdwioffset1;
 703     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 704     int              vdwioffset2;
 705     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 706     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 707     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 708     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 709     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 710     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 711     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 712     real             *charge;
 713     int              nvdwtype;
 714     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 715     int              *vdwtype;
 716     real             *vdwparam;
 717     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 718     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 719     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 720     real             rswitch_scalar,d_scalar;
 721     __m128           dummy_mask,cutoff_mask;
 722     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 723     __m128           one     = _mm_set1_ps(1.0);
 724     __m128           two     = _mm_set1_ps(2.0);
 725     x                = xx[0];
 726     f                = ff[0];
 727
 728     nri              = nlist->nri;
 729     iinr             = nlist->iinr;
 730     jindex           = nlist->jindex;
 731     jjnr             = nlist->jjnr;
 732     shiftidx         = nlist->shift;
 733     gid              = nlist->gid;
 734     shiftvec         = fr->shift_vec[0];
 735     fshift           = fr->fshift[0];
 736     facel            = _mm_set1_ps(fr->epsfac);
 737     charge           = mdatoms->chargeA;
 738     krf              = _mm_set1_ps(fr->ic->k_rf);
 739     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 740     crf              = _mm_set1_ps(fr->ic->c_rf);
 741     nvdwtype         = fr->ntype;
 742     vdwparam         = fr->nbfp;
 743     vdwtype          = mdatoms->typeA;
 744
 745     /* Setup water-specific parameters */
 746     inr              = nlist->iinr[0];
 747     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 748     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 749     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 750     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 751
 752     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 753     rcutoff_scalar   = fr->rcoulomb;
 754     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 755     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 756
 757     rswitch_scalar   = fr->rvdw_switch;
 758     rswitch          = _mm_set1_ps(rswitch_scalar);
 759     /* Setup switch parameters */
 760     d_scalar         = rcutoff_scalar-rswitch_scalar;
 761     d                = _mm_set1_ps(d_scalar);
 762     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 763     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 764     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 765     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 766     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 767     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 768
 769     /* Avoid stupid compiler warnings */
 770     jnrA = jnrB = jnrC = jnrD = 0;
 771     j_coord_offsetA = 0;
 772     j_coord_offsetB = 0;
 773     j_coord_offsetC = 0;
 774     j_coord_offsetD = 0;
 775
 776     outeriter        = 0;
 777     inneriter        = 0;
 778
 779     for(iidx=0;iidx<4*DIM;iidx++)
 780     {
 781         scratch[iidx] = 0.0;
 782     }
 783
 784     /* Start outer loop over neighborlists */
 785     for(iidx=0; iidx<nri; iidx++)
 786     {
 787         /* Load shift vector for this list */
 788         i_shift_offset   = DIM*shiftidx[iidx];
 789
 790         /* Load limits for loop over neighbors */
 791         j_index_start    = jindex[iidx];
 792         j_index_end      = jindex[iidx+1];
 793
 794         /* Get outer coordinate index */
 795         inr              = iinr[iidx];
 796         i_coord_offset   = DIM*inr;
 797
 798         /* Load i particle coords and add shift vector */
 799         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 800                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 801
 802         fix0             = _mm_setzero_ps();
 803         fiy0             = _mm_setzero_ps();
 804         fiz0             = _mm_setzero_ps();
 805         fix1             = _mm_setzero_ps();
 806         fiy1             = _mm_setzero_ps();
 807         fiz1             = _mm_setzero_ps();
 808         fix2             = _mm_setzero_ps();
 809         fiy2             = _mm_setzero_ps();
 810         fiz2             = _mm_setzero_ps();
 811
 812         /* Start inner kernel loop */
 813         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 814         {
 815
 816             /* Get j neighbor index, and coordinate index */
 817             jnrA             = jjnr[jidx];
 818             jnrB             = jjnr[jidx+1];
 819             jnrC             = jjnr[jidx+2];
 820             jnrD             = jjnr[jidx+3];
 821             j_coord_offsetA  = DIM*jnrA;
 822             j_coord_offsetB  = DIM*jnrB;
 823             j_coord_offsetC  = DIM*jnrC;
 824             j_coord_offsetD  = DIM*jnrD;
 825
 826             /* load j atom coordinates */
 827             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 828                                               x+j_coord_offsetC,x+j_coord_offsetD,
 829                                               &jx0,&jy0,&jz0);
 830
 831             /* Calculate displacement vector */
 832             dx00             = _mm_sub_ps(ix0,jx0);
 833             dy00             = _mm_sub_ps(iy0,jy0);
 834             dz00             = _mm_sub_ps(iz0,jz0);
 835             dx10             = _mm_sub_ps(ix1,jx0);
 836             dy10             = _mm_sub_ps(iy1,jy0);
 837             dz10             = _mm_sub_ps(iz1,jz0);
 838             dx20             = _mm_sub_ps(ix2,jx0);
 839             dy20             = _mm_sub_ps(iy2,jy0);
 840             dz20             = _mm_sub_ps(iz2,jz0);
 841
 842             /* Calculate squared distance and things based on it */
 843             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 844             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 845             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 846
 847             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 848             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 849             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 850
 851             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 852             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 853             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 854
 855             /* Load parameters for j particles */
 856             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 857                                                               charge+jnrC+0,charge+jnrD+0);
 858             vdwjidx0A        = 2*vdwtype[jnrA+0];
 859             vdwjidx0B        = 2*vdwtype[jnrB+0];
 860             vdwjidx0C        = 2*vdwtype[jnrC+0];
 861             vdwjidx0D        = 2*vdwtype[jnrD+0];
 862
 863             fjx0             = _mm_setzero_ps();
 864             fjy0             = _mm_setzero_ps();
 865             fjz0             = _mm_setzero_ps();
 866
 867             /**************************
 868              * CALCULATE INTERACTIONS *
 869              **************************/
 870
 871             if (gmx_mm_any_lt(rsq00,rcutoff2))
 872             {
 873
 874             r00              = _mm_mul_ps(rsq00,rinv00);
 875
 876             /* Compute parameters for interactions between i and j atoms */
 877             qq00             = _mm_mul_ps(iq0,jq0);
 878             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 879                                          vdwparam+vdwioffset0+vdwjidx0B,
 880                                          vdwparam+vdwioffset0+vdwjidx0C,
 881                                          vdwparam+vdwioffset0+vdwjidx0D,
 882                                          &c6_00,&c12_00);
 883
 884             /* REACTION-FIELD ELECTROSTATICS */
 885             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 886
 887             /* LENNARD-JONES DISPERSION/REPULSION */
 888
 889             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 890             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 891             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 892             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 893             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 894
 895             d                = _mm_sub_ps(r00,rswitch);
 896             d                = _mm_max_ps(d,_mm_setzero_ps());
 897             d2               = _mm_mul_ps(d,d);
 898             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 899
 900             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 901
 902             /* Evaluate switch function */
 903             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 904             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 905             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 906
 907             fscal            = _mm_add_ps(felec,fvdw);
 908
 909             fscal            = _mm_and_ps(fscal,cutoff_mask);
 910
 911             /* Calculate temporary vectorial force */
 912             tx               = _mm_mul_ps(fscal,dx00);
 913             ty               = _mm_mul_ps(fscal,dy00);
 914             tz               = _mm_mul_ps(fscal,dz00);
 915
 916             /* Update vectorial force */
 917             fix0             = _mm_add_ps(fix0,tx);
 918             fiy0             = _mm_add_ps(fiy0,ty);
 919             fiz0             = _mm_add_ps(fiz0,tz);
 920
 921             fjx0             = _mm_add_ps(fjx0,tx);
 922             fjy0             = _mm_add_ps(fjy0,ty);
 923             fjz0             = _mm_add_ps(fjz0,tz);
 924
 925             }
 926
 927             /**************************
 928              * CALCULATE INTERACTIONS *
 929              **************************/
 930
 931             if (gmx_mm_any_lt(rsq10,rcutoff2))
 932             {
 933
 934             /* Compute parameters for interactions between i and j atoms */
 935             qq10             = _mm_mul_ps(iq1,jq0);
 936
 937             /* REACTION-FIELD ELECTROSTATICS */
 938             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 939
 940             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 941
 942             fscal            = felec;
 943
 944             fscal            = _mm_and_ps(fscal,cutoff_mask);
 945
 946             /* Calculate temporary vectorial force */
 947             tx               = _mm_mul_ps(fscal,dx10);
 948             ty               = _mm_mul_ps(fscal,dy10);
 949             tz               = _mm_mul_ps(fscal,dz10);
 950
 951             /* Update vectorial force */
 952             fix1             = _mm_add_ps(fix1,tx);
 953             fiy1             = _mm_add_ps(fiy1,ty);
 954             fiz1             = _mm_add_ps(fiz1,tz);
 955
 956             fjx0             = _mm_add_ps(fjx0,tx);
 957             fjy0             = _mm_add_ps(fjy0,ty);
 958             fjz0             = _mm_add_ps(fjz0,tz);
 959
 960             }
 961
 962             /**************************
 963              * CALCULATE INTERACTIONS *
 964              **************************/
 965
 966             if (gmx_mm_any_lt(rsq20,rcutoff2))
 967             {
 968
 969             /* Compute parameters for interactions between i and j atoms */
 970             qq20             = _mm_mul_ps(iq2,jq0);
 971
 972             /* REACTION-FIELD ELECTROSTATICS */
 973             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 974
 975             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 976
 977             fscal            = felec;
 978
 979             fscal            = _mm_and_ps(fscal,cutoff_mask);
 980
 981             /* Calculate temporary vectorial force */
 982             tx               = _mm_mul_ps(fscal,dx20);
 983             ty               = _mm_mul_ps(fscal,dy20);
 984             tz               = _mm_mul_ps(fscal,dz20);
 985
 986             /* Update vectorial force */
 987             fix2             = _mm_add_ps(fix2,tx);
 988             fiy2             = _mm_add_ps(fiy2,ty);
 989             fiz2             = _mm_add_ps(fiz2,tz);
 990
 991             fjx0             = _mm_add_ps(fjx0,tx);
 992             fjy0             = _mm_add_ps(fjy0,ty);
 993             fjz0             = _mm_add_ps(fjz0,tz);
 994
 995             }
 996
 997             fjptrA             = f+j_coord_offsetA;
 998             fjptrB             = f+j_coord_offsetB;
 999             fjptrC             = f+j_coord_offsetC;
1000             fjptrD             = f+j_coord_offsetD;
1001
1002             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1003
1004             /* Inner loop uses 121 flops */
1005         }
1006
1007         if(jidx<j_index_end)
1008         {
1009
1010             /* Get j neighbor index, and coordinate index */
1011             jnrlistA         = jjnr[jidx];
1012             jnrlistB         = jjnr[jidx+1];
1013             jnrlistC         = jjnr[jidx+2];
1014             jnrlistD         = jjnr[jidx+3];
1015             /* Sign of each element will be negative for non-real atoms.
1016              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1017              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1018              */
1019             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1020             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1021             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1022             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1023             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1024             j_coord_offsetA  = DIM*jnrA;
1025             j_coord_offsetB  = DIM*jnrB;
1026             j_coord_offsetC  = DIM*jnrC;
1027             j_coord_offsetD  = DIM*jnrD;
1028
1029             /* load j atom coordinates */
1030             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1031                                               x+j_coord_offsetC,x+j_coord_offsetD,
1032                                               &jx0,&jy0,&jz0);
1033
1034             /* Calculate displacement vector */
1035             dx00             = _mm_sub_ps(ix0,jx0);
1036             dy00             = _mm_sub_ps(iy0,jy0);
1037             dz00             = _mm_sub_ps(iz0,jz0);
1038             dx10             = _mm_sub_ps(ix1,jx0);
1039             dy10             = _mm_sub_ps(iy1,jy0);
1040             dz10             = _mm_sub_ps(iz1,jz0);
1041             dx20             = _mm_sub_ps(ix2,jx0);
1042             dy20             = _mm_sub_ps(iy2,jy0);
1043             dz20             = _mm_sub_ps(iz2,jz0);
1044
1045             /* Calculate squared distance and things based on it */
1046             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1047             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1048             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1049
1050             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1051             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1052             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1053
1054             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1055             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1056             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1057
1058             /* Load parameters for j particles */
1059             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1060                                                               charge+jnrC+0,charge+jnrD+0);
1061             vdwjidx0A        = 2*vdwtype[jnrA+0];
1062             vdwjidx0B        = 2*vdwtype[jnrB+0];
1063             vdwjidx0C        = 2*vdwtype[jnrC+0];
1064             vdwjidx0D        = 2*vdwtype[jnrD+0];
1065
1066             fjx0             = _mm_setzero_ps();
1067             fjy0             = _mm_setzero_ps();
1068             fjz0             = _mm_setzero_ps();
1069
1070             /**************************
1071              * CALCULATE INTERACTIONS *
1072              **************************/
1073
1074             if (gmx_mm_any_lt(rsq00,rcutoff2))
1075             {
1076
1077             r00              = _mm_mul_ps(rsq00,rinv00);
1078             r00              = _mm_andnot_ps(dummy_mask,r00);
1079
1080             /* Compute parameters for interactions between i and j atoms */
1081             qq00             = _mm_mul_ps(iq0,jq0);
1082             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1083                                          vdwparam+vdwioffset0+vdwjidx0B,
1084                                          vdwparam+vdwioffset0+vdwjidx0C,
1085                                          vdwparam+vdwioffset0+vdwjidx0D,
1086                                          &c6_00,&c12_00);
1087
1088             /* REACTION-FIELD ELECTROSTATICS */
1089             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
1090
1091             /* LENNARD-JONES DISPERSION/REPULSION */
1092
1093             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1094             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
1095             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1096             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
1097             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1098
1099             d                = _mm_sub_ps(r00,rswitch);
1100             d                = _mm_max_ps(d,_mm_setzero_ps());
1101             d2               = _mm_mul_ps(d,d);
1102             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1103
1104             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1105
1106             /* Evaluate switch function */
1107             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1108             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1109             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1110
1111             fscal            = _mm_add_ps(felec,fvdw);
1112
1113             fscal            = _mm_and_ps(fscal,cutoff_mask);
1114
1115             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1116
1117             /* Calculate temporary vectorial force */
1118             tx               = _mm_mul_ps(fscal,dx00);
1119             ty               = _mm_mul_ps(fscal,dy00);
1120             tz               = _mm_mul_ps(fscal,dz00);
1121
1122             /* Update vectorial force */
1123             fix0             = _mm_add_ps(fix0,tx);
1124             fiy0             = _mm_add_ps(fiy0,ty);
1125             fiz0             = _mm_add_ps(fiz0,tz);
1126
1127             fjx0             = _mm_add_ps(fjx0,tx);
1128             fjy0             = _mm_add_ps(fjy0,ty);
1129             fjz0             = _mm_add_ps(fjz0,tz);
1130
1131             }
1132
1133             /**************************
1134              * CALCULATE INTERACTIONS *
1135              **************************/
1136
1137             if (gmx_mm_any_lt(rsq10,rcutoff2))
1138             {
1139
1140             /* Compute parameters for interactions between i and j atoms */
1141             qq10             = _mm_mul_ps(iq1,jq0);
1142
1143             /* REACTION-FIELD ELECTROSTATICS */
1144             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1145
1146             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1147
1148             fscal            = felec;
1149
1150             fscal            = _mm_and_ps(fscal,cutoff_mask);
1151
1152             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1153
1154             /* Calculate temporary vectorial force */
1155             tx               = _mm_mul_ps(fscal,dx10);
1156             ty               = _mm_mul_ps(fscal,dy10);
1157             tz               = _mm_mul_ps(fscal,dz10);
1158
1159             /* Update vectorial force */
1160             fix1             = _mm_add_ps(fix1,tx);
1161             fiy1             = _mm_add_ps(fiy1,ty);
1162             fiz1             = _mm_add_ps(fiz1,tz);
1163
1164             fjx0             = _mm_add_ps(fjx0,tx);
1165             fjy0             = _mm_add_ps(fjy0,ty);
1166             fjz0             = _mm_add_ps(fjz0,tz);
1167
1168             }
1169
1170             /**************************
1171              * CALCULATE INTERACTIONS *
1172              **************************/
1173
1174             if (gmx_mm_any_lt(rsq20,rcutoff2))
1175             {
1176
1177             /* Compute parameters for interactions between i and j atoms */
1178             qq20             = _mm_mul_ps(iq2,jq0);
1179
1180             /* REACTION-FIELD ELECTROSTATICS */
1181             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1182
1183             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1184
1185             fscal            = felec;
1186
1187             fscal            = _mm_and_ps(fscal,cutoff_mask);
1188
1189             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1190
1191             /* Calculate temporary vectorial force */
1192             tx               = _mm_mul_ps(fscal,dx20);
1193             ty               = _mm_mul_ps(fscal,dy20);
1194             tz               = _mm_mul_ps(fscal,dz20);
1195
1196             /* Update vectorial force */
1197             fix2             = _mm_add_ps(fix2,tx);
1198             fiy2             = _mm_add_ps(fiy2,ty);
1199             fiz2             = _mm_add_ps(fiz2,tz);
1200
1201             fjx0             = _mm_add_ps(fjx0,tx);
1202             fjy0             = _mm_add_ps(fjy0,ty);
1203             fjz0             = _mm_add_ps(fjz0,tz);
1204
1205             }
1206
1207             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1208             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1209             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1210             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1211
1212             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1213
1214             /* Inner loop uses 122 flops */
1215         }
1216
1217         /* End of innermost loop */
1218
1219         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1220                                               f+i_coord_offset,fshift+i_shift_offset);
1221
1222         /* Increment number of inner iterations */
1223         inneriter                  += j_index_end - j_index_start;
1224
1225         /* Outer loop uses 18 flops */
1226     }
1227
1228     /* Increment number of outer iterations */
1229     outeriter        += nri;
1230
1231     /* Update outer/inner flops */
1232
1233     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*122);
1234 }