src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_sse4_1_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse4_1_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_VF_sse4_1_single
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_VF_sse4_1_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  89     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 100     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 101     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 102     real             rswitch_scalar,d_scalar;
 103     __m128           dummy_mask,cutoff_mask;
 104     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 105     __m128           one     = _mm_set1_ps(1.0);
 106     __m128           two     = _mm_set1_ps(2.0);
 107     x                = xx[0];
 108     f                = ff[0];
 109
 110     nri              = nlist->nri;
 111     iinr             = nlist->iinr;
 112     jindex           = nlist->jindex;
 113     jjnr             = nlist->jjnr;
 114     shiftidx         = nlist->shift;
 115     gid              = nlist->gid;
 116     shiftvec         = fr->shift_vec[0];
 117     fshift           = fr->fshift[0];
 118     facel            = _mm_set1_ps(fr->ic->epsfac);
 119     charge           = mdatoms->chargeA;
 120     krf              = _mm_set1_ps(fr->ic->k_rf);
 121     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 122     crf              = _mm_set1_ps(fr->ic->c_rf);
 123     nvdwtype         = fr->ntype;
 124     vdwparam         = fr->nbfp;
 125     vdwtype          = mdatoms->typeA;
 126
 127     /* Setup water-specific parameters */
 128     inr              = nlist->iinr[0];
 129     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 130     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 131     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 132     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 133
 134     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 135     rcutoff_scalar   = fr->ic->rcoulomb;
 136     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 137     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 138
 139     rswitch_scalar   = fr->ic->rvdw_switch;
 140     rswitch          = _mm_set1_ps(rswitch_scalar);
 141     /* Setup switch parameters */
 142     d_scalar         = rcutoff_scalar-rswitch_scalar;
 143     d                = _mm_set1_ps(d_scalar);
 144     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 145     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 146     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 147     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 148     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 149     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 150
 151     /* Avoid stupid compiler warnings */
 152     jnrA = jnrB = jnrC = jnrD = 0;
 153     j_coord_offsetA = 0;
 154     j_coord_offsetB = 0;
 155     j_coord_offsetC = 0;
 156     j_coord_offsetD = 0;
 157
 158     outeriter        = 0;
 159     inneriter        = 0;
 160
 161     for(iidx=0;iidx<4*DIM;iidx++)
 162     {
 163         scratch[iidx] = 0.0;
 164     }
 165
 166     /* Start outer loop over neighborlists */
 167     for(iidx=0; iidx<nri; iidx++)
 168     {
 169         /* Load shift vector for this list */
 170         i_shift_offset   = DIM*shiftidx[iidx];
 171
 172         /* Load limits for loop over neighbors */
 173         j_index_start    = jindex[iidx];
 174         j_index_end      = jindex[iidx+1];
 175
 176         /* Get outer coordinate index */
 177         inr              = iinr[iidx];
 178         i_coord_offset   = DIM*inr;
 179
 180         /* Load i particle coords and add shift vector */
 181         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 182                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 183
 184         fix0             = _mm_setzero_ps();
 185         fiy0             = _mm_setzero_ps();
 186         fiz0             = _mm_setzero_ps();
 187         fix1             = _mm_setzero_ps();
 188         fiy1             = _mm_setzero_ps();
 189         fiz1             = _mm_setzero_ps();
 190         fix2             = _mm_setzero_ps();
 191         fiy2             = _mm_setzero_ps();
 192         fiz2             = _mm_setzero_ps();
 193
 194         /* Reset potential sums */
 195         velecsum         = _mm_setzero_ps();
 196         vvdwsum          = _mm_setzero_ps();
 197
 198         /* Start inner kernel loop */
 199         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 200         {
 201
 202             /* Get j neighbor index, and coordinate index */
 203             jnrA             = jjnr[jidx];
 204             jnrB             = jjnr[jidx+1];
 205             jnrC             = jjnr[jidx+2];
 206             jnrD             = jjnr[jidx+3];
 207             j_coord_offsetA  = DIM*jnrA;
 208             j_coord_offsetB  = DIM*jnrB;
 209             j_coord_offsetC  = DIM*jnrC;
 210             j_coord_offsetD  = DIM*jnrD;
 211
 212             /* load j atom coordinates */
 213             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 214                                               x+j_coord_offsetC,x+j_coord_offsetD,
 215                                               &jx0,&jy0,&jz0);
 216
 217             /* Calculate displacement vector */
 218             dx00             = _mm_sub_ps(ix0,jx0);
 219             dy00             = _mm_sub_ps(iy0,jy0);
 220             dz00             = _mm_sub_ps(iz0,jz0);
 221             dx10             = _mm_sub_ps(ix1,jx0);
 222             dy10             = _mm_sub_ps(iy1,jy0);
 223             dz10             = _mm_sub_ps(iz1,jz0);
 224             dx20             = _mm_sub_ps(ix2,jx0);
 225             dy20             = _mm_sub_ps(iy2,jy0);
 226             dz20             = _mm_sub_ps(iz2,jz0);
 227
 228             /* Calculate squared distance and things based on it */
 229             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 230             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 231             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 232
 233             rinv00           = sse41_invsqrt_f(rsq00);
 234             rinv10           = sse41_invsqrt_f(rsq10);
 235             rinv20           = sse41_invsqrt_f(rsq20);
 236
 237             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 238             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 239             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 240
 241             /* Load parameters for j particles */
 242             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 243                                                               charge+jnrC+0,charge+jnrD+0);
 244             vdwjidx0A        = 2*vdwtype[jnrA+0];
 245             vdwjidx0B        = 2*vdwtype[jnrB+0];
 246             vdwjidx0C        = 2*vdwtype[jnrC+0];
 247             vdwjidx0D        = 2*vdwtype[jnrD+0];
 248
 249             fjx0             = _mm_setzero_ps();
 250             fjy0             = _mm_setzero_ps();
 251             fjz0             = _mm_setzero_ps();
 252
 253             /**************************
 254              * CALCULATE INTERACTIONS *
 255              **************************/
 256
 257             if (gmx_mm_any_lt(rsq00,rcutoff2))
 258             {
 259
 260             r00              = _mm_mul_ps(rsq00,rinv00);
 261
 262             /* Compute parameters for interactions between i and j atoms */
 263             qq00             = _mm_mul_ps(iq0,jq0);
 264             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 265                                          vdwparam+vdwioffset0+vdwjidx0B,
 266                                          vdwparam+vdwioffset0+vdwjidx0C,
 267                                          vdwparam+vdwioffset0+vdwjidx0D,
 268                                          &c6_00,&c12_00);
 269
 270             /* REACTION-FIELD ELECTROSTATICS */
 271             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 272             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 273
 274             /* LENNARD-JONES DISPERSION/REPULSION */
 275
 276             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 277             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 278             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 279             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 280             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 281
 282             d                = _mm_sub_ps(r00,rswitch);
 283             d                = _mm_max_ps(d,_mm_setzero_ps());
 284             d2               = _mm_mul_ps(d,d);
 285             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)))))));
 286
 287             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 288
 289             /* Evaluate switch function */
 290             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 291             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 292             vvdw             = _mm_mul_ps(vvdw,sw);
 293             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 294
 295             /* Update potential sum for this i atom from the interaction with this j atom. */
 296             velec            = _mm_and_ps(velec,cutoff_mask);
 297             velecsum         = _mm_add_ps(velecsum,velec);
 298             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 299             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 300
 301             fscal            = _mm_add_ps(felec,fvdw);
 302
 303             fscal            = _mm_and_ps(fscal,cutoff_mask);
 304
 305             /* Calculate temporary vectorial force */
 306             tx               = _mm_mul_ps(fscal,dx00);
 307             ty               = _mm_mul_ps(fscal,dy00);
 308             tz               = _mm_mul_ps(fscal,dz00);
 309
 310             /* Update vectorial force */
 311             fix0             = _mm_add_ps(fix0,tx);
 312             fiy0             = _mm_add_ps(fiy0,ty);
 313             fiz0             = _mm_add_ps(fiz0,tz);
 314
 315             fjx0             = _mm_add_ps(fjx0,tx);
 316             fjy0             = _mm_add_ps(fjy0,ty);
 317             fjz0             = _mm_add_ps(fjz0,tz);
 318
 319             }
 320
 321             /**************************
 322              * CALCULATE INTERACTIONS *
 323              **************************/
 324
 325             if (gmx_mm_any_lt(rsq10,rcutoff2))
 326             {
 327
 328             /* Compute parameters for interactions between i and j atoms */
 329             qq10             = _mm_mul_ps(iq1,jq0);
 330
 331             /* REACTION-FIELD ELECTROSTATICS */
 332             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 333             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 334
 335             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 336
 337             /* Update potential sum for this i atom from the interaction with this j atom. */
 338             velec            = _mm_and_ps(velec,cutoff_mask);
 339             velecsum         = _mm_add_ps(velecsum,velec);
 340
 341             fscal            = felec;
 342
 343             fscal            = _mm_and_ps(fscal,cutoff_mask);
 344
 345             /* Calculate temporary vectorial force */
 346             tx               = _mm_mul_ps(fscal,dx10);
 347             ty               = _mm_mul_ps(fscal,dy10);
 348             tz               = _mm_mul_ps(fscal,dz10);
 349
 350             /* Update vectorial force */
 351             fix1             = _mm_add_ps(fix1,tx);
 352             fiy1             = _mm_add_ps(fiy1,ty);
 353             fiz1             = _mm_add_ps(fiz1,tz);
 354
 355             fjx0             = _mm_add_ps(fjx0,tx);
 356             fjy0             = _mm_add_ps(fjy0,ty);
 357             fjz0             = _mm_add_ps(fjz0,tz);
 358
 359             }
 360
 361             /**************************
 362              * CALCULATE INTERACTIONS *
 363              **************************/
 364
 365             if (gmx_mm_any_lt(rsq20,rcutoff2))
 366             {
 367
 368             /* Compute parameters for interactions between i and j atoms */
 369             qq20             = _mm_mul_ps(iq2,jq0);
 370
 371             /* REACTION-FIELD ELECTROSTATICS */
 372             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 373             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 374
 375             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 376
 377             /* Update potential sum for this i atom from the interaction with this j atom. */
 378             velec            = _mm_and_ps(velec,cutoff_mask);
 379             velecsum         = _mm_add_ps(velecsum,velec);
 380
 381             fscal            = felec;
 382
 383             fscal            = _mm_and_ps(fscal,cutoff_mask);
 384
 385             /* Calculate temporary vectorial force */
 386             tx               = _mm_mul_ps(fscal,dx20);
 387             ty               = _mm_mul_ps(fscal,dy20);
 388             tz               = _mm_mul_ps(fscal,dz20);
 389
 390             /* Update vectorial force */
 391             fix2             = _mm_add_ps(fix2,tx);
 392             fiy2             = _mm_add_ps(fiy2,ty);
 393             fiz2             = _mm_add_ps(fiz2,tz);
 394
 395             fjx0             = _mm_add_ps(fjx0,tx);
 396             fjy0             = _mm_add_ps(fjy0,ty);
 397             fjz0             = _mm_add_ps(fjz0,tz);
 398
 399             }
 400
 401             fjptrA             = f+j_coord_offsetA;
 402             fjptrB             = f+j_coord_offsetB;
 403             fjptrC             = f+j_coord_offsetC;
 404             fjptrD             = f+j_coord_offsetD;
 405
 406             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 407
 408             /* Inner loop uses 142 flops */
 409         }
 410
 411         if(jidx<j_index_end)
 412         {
 413
 414             /* Get j neighbor index, and coordinate index */
 415             jnrlistA         = jjnr[jidx];
 416             jnrlistB         = jjnr[jidx+1];
 417             jnrlistC         = jjnr[jidx+2];
 418             jnrlistD         = jjnr[jidx+3];
 419             /* Sign of each element will be negative for non-real atoms.
 420              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 421              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 422              */
 423             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 424             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 425             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 426             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 427             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 428             j_coord_offsetA  = DIM*jnrA;
 429             j_coord_offsetB  = DIM*jnrB;
 430             j_coord_offsetC  = DIM*jnrC;
 431             j_coord_offsetD  = DIM*jnrD;
 432
 433             /* load j atom coordinates */
 434             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 435                                               x+j_coord_offsetC,x+j_coord_offsetD,
 436                                               &jx0,&jy0,&jz0);
 437
 438             /* Calculate displacement vector */
 439             dx00             = _mm_sub_ps(ix0,jx0);
 440             dy00             = _mm_sub_ps(iy0,jy0);
 441             dz00             = _mm_sub_ps(iz0,jz0);
 442             dx10             = _mm_sub_ps(ix1,jx0);
 443             dy10             = _mm_sub_ps(iy1,jy0);
 444             dz10             = _mm_sub_ps(iz1,jz0);
 445             dx20             = _mm_sub_ps(ix2,jx0);
 446             dy20             = _mm_sub_ps(iy2,jy0);
 447             dz20             = _mm_sub_ps(iz2,jz0);
 448
 449             /* Calculate squared distance and things based on it */
 450             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 451             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 452             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 453
 454             rinv00           = sse41_invsqrt_f(rsq00);
 455             rinv10           = sse41_invsqrt_f(rsq10);
 456             rinv20           = sse41_invsqrt_f(rsq20);
 457
 458             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 459             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 460             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 461
 462             /* Load parameters for j particles */
 463             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 464                                                               charge+jnrC+0,charge+jnrD+0);
 465             vdwjidx0A        = 2*vdwtype[jnrA+0];
 466             vdwjidx0B        = 2*vdwtype[jnrB+0];
 467             vdwjidx0C        = 2*vdwtype[jnrC+0];
 468             vdwjidx0D        = 2*vdwtype[jnrD+0];
 469
 470             fjx0             = _mm_setzero_ps();
 471             fjy0             = _mm_setzero_ps();
 472             fjz0             = _mm_setzero_ps();
 473
 474             /**************************
 475              * CALCULATE INTERACTIONS *
 476              **************************/
 477
 478             if (gmx_mm_any_lt(rsq00,rcutoff2))
 479             {
 480
 481             r00              = _mm_mul_ps(rsq00,rinv00);
 482             r00              = _mm_andnot_ps(dummy_mask,r00);
 483
 484             /* Compute parameters for interactions between i and j atoms */
 485             qq00             = _mm_mul_ps(iq0,jq0);
 486             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 487                                          vdwparam+vdwioffset0+vdwjidx0B,
 488                                          vdwparam+vdwioffset0+vdwjidx0C,
 489                                          vdwparam+vdwioffset0+vdwjidx0D,
 490                                          &c6_00,&c12_00);
 491
 492             /* REACTION-FIELD ELECTROSTATICS */
 493             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 494             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 495
 496             /* LENNARD-JONES DISPERSION/REPULSION */
 497
 498             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 499             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 500             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 501             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 502             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 503
 504             d                = _mm_sub_ps(r00,rswitch);
 505             d                = _mm_max_ps(d,_mm_setzero_ps());
 506             d2               = _mm_mul_ps(d,d);
 507             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)))))));
 508
 509             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 510
 511             /* Evaluate switch function */
 512             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 513             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 514             vvdw             = _mm_mul_ps(vvdw,sw);
 515             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 516
 517             /* Update potential sum for this i atom from the interaction with this j atom. */
 518             velec            = _mm_and_ps(velec,cutoff_mask);
 519             velec            = _mm_andnot_ps(dummy_mask,velec);
 520             velecsum         = _mm_add_ps(velecsum,velec);
 521             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 522             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 523             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 524
 525             fscal            = _mm_add_ps(felec,fvdw);
 526
 527             fscal            = _mm_and_ps(fscal,cutoff_mask);
 528
 529             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 530
 531             /* Calculate temporary vectorial force */
 532             tx               = _mm_mul_ps(fscal,dx00);
 533             ty               = _mm_mul_ps(fscal,dy00);
 534             tz               = _mm_mul_ps(fscal,dz00);
 535
 536             /* Update vectorial force */
 537             fix0             = _mm_add_ps(fix0,tx);
 538             fiy0             = _mm_add_ps(fiy0,ty);
 539             fiz0             = _mm_add_ps(fiz0,tz);
 540
 541             fjx0             = _mm_add_ps(fjx0,tx);
 542             fjy0             = _mm_add_ps(fjy0,ty);
 543             fjz0             = _mm_add_ps(fjz0,tz);
 544
 545             }
 546
 547             /**************************
 548              * CALCULATE INTERACTIONS *
 549              **************************/
 550
 551             if (gmx_mm_any_lt(rsq10,rcutoff2))
 552             {
 553
 554             /* Compute parameters for interactions between i and j atoms */
 555             qq10             = _mm_mul_ps(iq1,jq0);
 556
 557             /* REACTION-FIELD ELECTROSTATICS */
 558             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 559             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 560
 561             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 562
 563             /* Update potential sum for this i atom from the interaction with this j atom. */
 564             velec            = _mm_and_ps(velec,cutoff_mask);
 565             velec            = _mm_andnot_ps(dummy_mask,velec);
 566             velecsum         = _mm_add_ps(velecsum,velec);
 567
 568             fscal            = felec;
 569
 570             fscal            = _mm_and_ps(fscal,cutoff_mask);
 571
 572             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 573
 574             /* Calculate temporary vectorial force */
 575             tx               = _mm_mul_ps(fscal,dx10);
 576             ty               = _mm_mul_ps(fscal,dy10);
 577             tz               = _mm_mul_ps(fscal,dz10);
 578
 579             /* Update vectorial force */
 580             fix1             = _mm_add_ps(fix1,tx);
 581             fiy1             = _mm_add_ps(fiy1,ty);
 582             fiz1             = _mm_add_ps(fiz1,tz);
 583
 584             fjx0             = _mm_add_ps(fjx0,tx);
 585             fjy0             = _mm_add_ps(fjy0,ty);
 586             fjz0             = _mm_add_ps(fjz0,tz);
 587
 588             }
 589
 590             /**************************
 591              * CALCULATE INTERACTIONS *
 592              **************************/
 593
 594             if (gmx_mm_any_lt(rsq20,rcutoff2))
 595             {
 596
 597             /* Compute parameters for interactions between i and j atoms */
 598             qq20             = _mm_mul_ps(iq2,jq0);
 599
 600             /* REACTION-FIELD ELECTROSTATICS */
 601             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 602             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 603
 604             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 605
 606             /* Update potential sum for this i atom from the interaction with this j atom. */
 607             velec            = _mm_and_ps(velec,cutoff_mask);
 608             velec            = _mm_andnot_ps(dummy_mask,velec);
 609             velecsum         = _mm_add_ps(velecsum,velec);
 610
 611             fscal            = felec;
 612
 613             fscal            = _mm_and_ps(fscal,cutoff_mask);
 614
 615             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 616
 617             /* Calculate temporary vectorial force */
 618             tx               = _mm_mul_ps(fscal,dx20);
 619             ty               = _mm_mul_ps(fscal,dy20);
 620             tz               = _mm_mul_ps(fscal,dz20);
 621
 622             /* Update vectorial force */
 623             fix2             = _mm_add_ps(fix2,tx);
 624             fiy2             = _mm_add_ps(fiy2,ty);
 625             fiz2             = _mm_add_ps(fiz2,tz);
 626
 627             fjx0             = _mm_add_ps(fjx0,tx);
 628             fjy0             = _mm_add_ps(fjy0,ty);
 629             fjz0             = _mm_add_ps(fjz0,tz);
 630
 631             }
 632
 633             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 634             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 635             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 636             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 637
 638             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 639
 640             /* Inner loop uses 143 flops */
 641         }
 642
 643         /* End of innermost loop */
 644
 645         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 646                                               f+i_coord_offset,fshift+i_shift_offset);
 647
 648         ggid                        = gid[iidx];
 649         /* Update potential energies */
 650         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 651         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 652
 653         /* Increment number of inner iterations */
 654         inneriter                  += j_index_end - j_index_start;
 655
 656         /* Outer loop uses 20 flops */
 657     }
 658
 659     /* Increment number of outer iterations */
 660     outeriter        += nri;
 661
 662     /* Update outer/inner flops */
 663
 664     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*143);
 665 }
 666 /*
 667  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_sse4_1_single
 668  * Electrostatics interaction: ReactionField
 669  * VdW interaction:            LennardJones
 670  * Geometry:                   Water3-Particle
 671  * Calculate force/pot:        Force
 672  */
 673 void
 674 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_sse4_1_single
 675                     (t_nblist                    * gmx_restrict       nlist,
 676                      rvec                        * gmx_restrict          xx,
 677                      rvec                        * gmx_restrict          ff,
 678                      struct t_forcerec           * gmx_restrict          fr,
 679                      t_mdatoms                   * gmx_restrict     mdatoms,
 680                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 681                      t_nrnb                      * gmx_restrict        nrnb)
 682 {
 683     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 684      * just 0 for non-waters.
 685      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 686      * jnr indices corresponding to data put in the four positions in the SIMD register.
 687      */
 688     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 689     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 690     int              jnrA,jnrB,jnrC,jnrD;
 691     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 692     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 693     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 694     real             rcutoff_scalar;
 695     real             *shiftvec,*fshift,*x,*f;
 696     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 697     real             scratch[4*DIM];
 698     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 699     int              vdwioffset0;
 700     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 701     int              vdwioffset1;
 702     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 703     int              vdwioffset2;
 704     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 705     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 706     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 707     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 708     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 709     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 710     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 711     real             *charge;
 712     int              nvdwtype;
 713     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 714     int              *vdwtype;
 715     real             *vdwparam;
 716     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 717     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 718     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 719     real             rswitch_scalar,d_scalar;
 720     __m128           dummy_mask,cutoff_mask;
 721     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 722     __m128           one     = _mm_set1_ps(1.0);
 723     __m128           two     = _mm_set1_ps(2.0);
 724     x                = xx[0];
 725     f                = ff[0];
 726
 727     nri              = nlist->nri;
 728     iinr             = nlist->iinr;
 729     jindex           = nlist->jindex;
 730     jjnr             = nlist->jjnr;
 731     shiftidx         = nlist->shift;
 732     gid              = nlist->gid;
 733     shiftvec         = fr->shift_vec[0];
 734     fshift           = fr->fshift[0];
 735     facel            = _mm_set1_ps(fr->ic->epsfac);
 736     charge           = mdatoms->chargeA;
 737     krf              = _mm_set1_ps(fr->ic->k_rf);
 738     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 739     crf              = _mm_set1_ps(fr->ic->c_rf);
 740     nvdwtype         = fr->ntype;
 741     vdwparam         = fr->nbfp;
 742     vdwtype          = mdatoms->typeA;
 743
 744     /* Setup water-specific parameters */
 745     inr              = nlist->iinr[0];
 746     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 747     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 748     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 749     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 750
 751     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 752     rcutoff_scalar   = fr->ic->rcoulomb;
 753     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 754     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 755
 756     rswitch_scalar   = fr->ic->rvdw_switch;
 757     rswitch          = _mm_set1_ps(rswitch_scalar);
 758     /* Setup switch parameters */
 759     d_scalar         = rcutoff_scalar-rswitch_scalar;
 760     d                = _mm_set1_ps(d_scalar);
 761     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 762     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 763     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 764     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 765     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 766     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 767
 768     /* Avoid stupid compiler warnings */
 769     jnrA = jnrB = jnrC = jnrD = 0;
 770     j_coord_offsetA = 0;
 771     j_coord_offsetB = 0;
 772     j_coord_offsetC = 0;
 773     j_coord_offsetD = 0;
 774
 775     outeriter        = 0;
 776     inneriter        = 0;
 777
 778     for(iidx=0;iidx<4*DIM;iidx++)
 779     {
 780         scratch[iidx] = 0.0;
 781     }
 782
 783     /* Start outer loop over neighborlists */
 784     for(iidx=0; iidx<nri; iidx++)
 785     {
 786         /* Load shift vector for this list */
 787         i_shift_offset   = DIM*shiftidx[iidx];
 788
 789         /* Load limits for loop over neighbors */
 790         j_index_start    = jindex[iidx];
 791         j_index_end      = jindex[iidx+1];
 792
 793         /* Get outer coordinate index */
 794         inr              = iinr[iidx];
 795         i_coord_offset   = DIM*inr;
 796
 797         /* Load i particle coords and add shift vector */
 798         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 799                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 800
 801         fix0             = _mm_setzero_ps();
 802         fiy0             = _mm_setzero_ps();
 803         fiz0             = _mm_setzero_ps();
 804         fix1             = _mm_setzero_ps();
 805         fiy1             = _mm_setzero_ps();
 806         fiz1             = _mm_setzero_ps();
 807         fix2             = _mm_setzero_ps();
 808         fiy2             = _mm_setzero_ps();
 809         fiz2             = _mm_setzero_ps();
 810
 811         /* Start inner kernel loop */
 812         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 813         {
 814
 815             /* Get j neighbor index, and coordinate index */
 816             jnrA             = jjnr[jidx];
 817             jnrB             = jjnr[jidx+1];
 818             jnrC             = jjnr[jidx+2];
 819             jnrD             = jjnr[jidx+3];
 820             j_coord_offsetA  = DIM*jnrA;
 821             j_coord_offsetB  = DIM*jnrB;
 822             j_coord_offsetC  = DIM*jnrC;
 823             j_coord_offsetD  = DIM*jnrD;
 824
 825             /* load j atom coordinates */
 826             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 827                                               x+j_coord_offsetC,x+j_coord_offsetD,
 828                                               &jx0,&jy0,&jz0);
 829
 830             /* Calculate displacement vector */
 831             dx00             = _mm_sub_ps(ix0,jx0);
 832             dy00             = _mm_sub_ps(iy0,jy0);
 833             dz00             = _mm_sub_ps(iz0,jz0);
 834             dx10             = _mm_sub_ps(ix1,jx0);
 835             dy10             = _mm_sub_ps(iy1,jy0);
 836             dz10             = _mm_sub_ps(iz1,jz0);
 837             dx20             = _mm_sub_ps(ix2,jx0);
 838             dy20             = _mm_sub_ps(iy2,jy0);
 839             dz20             = _mm_sub_ps(iz2,jz0);
 840
 841             /* Calculate squared distance and things based on it */
 842             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 843             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 844             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 845
 846             rinv00           = sse41_invsqrt_f(rsq00);
 847             rinv10           = sse41_invsqrt_f(rsq10);
 848             rinv20           = sse41_invsqrt_f(rsq20);
 849
 850             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 851             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 852             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 853
 854             /* Load parameters for j particles */
 855             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 856                                                               charge+jnrC+0,charge+jnrD+0);
 857             vdwjidx0A        = 2*vdwtype[jnrA+0];
 858             vdwjidx0B        = 2*vdwtype[jnrB+0];
 859             vdwjidx0C        = 2*vdwtype[jnrC+0];
 860             vdwjidx0D        = 2*vdwtype[jnrD+0];
 861
 862             fjx0             = _mm_setzero_ps();
 863             fjy0             = _mm_setzero_ps();
 864             fjz0             = _mm_setzero_ps();
 865
 866             /**************************
 867              * CALCULATE INTERACTIONS *
 868              **************************/
 869
 870             if (gmx_mm_any_lt(rsq00,rcutoff2))
 871             {
 872
 873             r00              = _mm_mul_ps(rsq00,rinv00);
 874
 875             /* Compute parameters for interactions between i and j atoms */
 876             qq00             = _mm_mul_ps(iq0,jq0);
 877             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 878                                          vdwparam+vdwioffset0+vdwjidx0B,
 879                                          vdwparam+vdwioffset0+vdwjidx0C,
 880                                          vdwparam+vdwioffset0+vdwjidx0D,
 881                                          &c6_00,&c12_00);
 882
 883             /* REACTION-FIELD ELECTROSTATICS */
 884             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 885
 886             /* LENNARD-JONES DISPERSION/REPULSION */
 887
 888             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 889             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 890             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 891             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 892             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 893
 894             d                = _mm_sub_ps(r00,rswitch);
 895             d                = _mm_max_ps(d,_mm_setzero_ps());
 896             d2               = _mm_mul_ps(d,d);
 897             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)))))));
 898
 899             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 900
 901             /* Evaluate switch function */
 902             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 903             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 904             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 905
 906             fscal            = _mm_add_ps(felec,fvdw);
 907
 908             fscal            = _mm_and_ps(fscal,cutoff_mask);
 909
 910             /* Calculate temporary vectorial force */
 911             tx               = _mm_mul_ps(fscal,dx00);
 912             ty               = _mm_mul_ps(fscal,dy00);
 913             tz               = _mm_mul_ps(fscal,dz00);
 914
 915             /* Update vectorial force */
 916             fix0             = _mm_add_ps(fix0,tx);
 917             fiy0             = _mm_add_ps(fiy0,ty);
 918             fiz0             = _mm_add_ps(fiz0,tz);
 919
 920             fjx0             = _mm_add_ps(fjx0,tx);
 921             fjy0             = _mm_add_ps(fjy0,ty);
 922             fjz0             = _mm_add_ps(fjz0,tz);
 923
 924             }
 925
 926             /**************************
 927              * CALCULATE INTERACTIONS *
 928              **************************/
 929
 930             if (gmx_mm_any_lt(rsq10,rcutoff2))
 931             {
 932
 933             /* Compute parameters for interactions between i and j atoms */
 934             qq10             = _mm_mul_ps(iq1,jq0);
 935
 936             /* REACTION-FIELD ELECTROSTATICS */
 937             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 938
 939             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 940
 941             fscal            = felec;
 942
 943             fscal            = _mm_and_ps(fscal,cutoff_mask);
 944
 945             /* Calculate temporary vectorial force */
 946             tx               = _mm_mul_ps(fscal,dx10);
 947             ty               = _mm_mul_ps(fscal,dy10);
 948             tz               = _mm_mul_ps(fscal,dz10);
 949
 950             /* Update vectorial force */
 951             fix1             = _mm_add_ps(fix1,tx);
 952             fiy1             = _mm_add_ps(fiy1,ty);
 953             fiz1             = _mm_add_ps(fiz1,tz);
 954
 955             fjx0             = _mm_add_ps(fjx0,tx);
 956             fjy0             = _mm_add_ps(fjy0,ty);
 957             fjz0             = _mm_add_ps(fjz0,tz);
 958
 959             }
 960
 961             /**************************
 962              * CALCULATE INTERACTIONS *
 963              **************************/
 964
 965             if (gmx_mm_any_lt(rsq20,rcutoff2))
 966             {
 967
 968             /* Compute parameters for interactions between i and j atoms */
 969             qq20             = _mm_mul_ps(iq2,jq0);
 970
 971             /* REACTION-FIELD ELECTROSTATICS */
 972             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 973
 974             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 975
 976             fscal            = felec;
 977
 978             fscal            = _mm_and_ps(fscal,cutoff_mask);
 979
 980             /* Calculate temporary vectorial force */
 981             tx               = _mm_mul_ps(fscal,dx20);
 982             ty               = _mm_mul_ps(fscal,dy20);
 983             tz               = _mm_mul_ps(fscal,dz20);
 984
 985             /* Update vectorial force */
 986             fix2             = _mm_add_ps(fix2,tx);
 987             fiy2             = _mm_add_ps(fiy2,ty);
 988             fiz2             = _mm_add_ps(fiz2,tz);
 989
 990             fjx0             = _mm_add_ps(fjx0,tx);
 991             fjy0             = _mm_add_ps(fjy0,ty);
 992             fjz0             = _mm_add_ps(fjz0,tz);
 993
 994             }
 995
 996             fjptrA             = f+j_coord_offsetA;
 997             fjptrB             = f+j_coord_offsetB;
 998             fjptrC             = f+j_coord_offsetC;
 999             fjptrD             = f+j_coord_offsetD;
1000
1001             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1002
1003             /* Inner loop uses 121 flops */
1004         }
1005
1006         if(jidx<j_index_end)
1007         {
1008
1009             /* Get j neighbor index, and coordinate index */
1010             jnrlistA         = jjnr[jidx];
1011             jnrlistB         = jjnr[jidx+1];
1012             jnrlistC         = jjnr[jidx+2];
1013             jnrlistD         = jjnr[jidx+3];
1014             /* Sign of each element will be negative for non-real atoms.
1015              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1016              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1017              */
1018             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1019             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1020             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1021             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1022             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1023             j_coord_offsetA  = DIM*jnrA;
1024             j_coord_offsetB  = DIM*jnrB;
1025             j_coord_offsetC  = DIM*jnrC;
1026             j_coord_offsetD  = DIM*jnrD;
1027
1028             /* load j atom coordinates */
1029             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1030                                               x+j_coord_offsetC,x+j_coord_offsetD,
1031                                               &jx0,&jy0,&jz0);
1032
1033             /* Calculate displacement vector */
1034             dx00             = _mm_sub_ps(ix0,jx0);
1035             dy00             = _mm_sub_ps(iy0,jy0);
1036             dz00             = _mm_sub_ps(iz0,jz0);
1037             dx10             = _mm_sub_ps(ix1,jx0);
1038             dy10             = _mm_sub_ps(iy1,jy0);
1039             dz10             = _mm_sub_ps(iz1,jz0);
1040             dx20             = _mm_sub_ps(ix2,jx0);
1041             dy20             = _mm_sub_ps(iy2,jy0);
1042             dz20             = _mm_sub_ps(iz2,jz0);
1043
1044             /* Calculate squared distance and things based on it */
1045             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1046             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1047             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1048
1049             rinv00           = sse41_invsqrt_f(rsq00);
1050             rinv10           = sse41_invsqrt_f(rsq10);
1051             rinv20           = sse41_invsqrt_f(rsq20);
1052
1053             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1054             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1055             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1056
1057             /* Load parameters for j particles */
1058             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1059                                                               charge+jnrC+0,charge+jnrD+0);
1060             vdwjidx0A        = 2*vdwtype[jnrA+0];
1061             vdwjidx0B        = 2*vdwtype[jnrB+0];
1062             vdwjidx0C        = 2*vdwtype[jnrC+0];
1063             vdwjidx0D        = 2*vdwtype[jnrD+0];
1064
1065             fjx0             = _mm_setzero_ps();
1066             fjy0             = _mm_setzero_ps();
1067             fjz0             = _mm_setzero_ps();
1068
1069             /**************************
1070              * CALCULATE INTERACTIONS *
1071              **************************/
1072
1073             if (gmx_mm_any_lt(rsq00,rcutoff2))
1074             {
1075
1076             r00              = _mm_mul_ps(rsq00,rinv00);
1077             r00              = _mm_andnot_ps(dummy_mask,r00);
1078
1079             /* Compute parameters for interactions between i and j atoms */
1080             qq00             = _mm_mul_ps(iq0,jq0);
1081             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1082                                          vdwparam+vdwioffset0+vdwjidx0B,
1083                                          vdwparam+vdwioffset0+vdwjidx0C,
1084                                          vdwparam+vdwioffset0+vdwjidx0D,
1085                                          &c6_00,&c12_00);
1086
1087             /* REACTION-FIELD ELECTROSTATICS */
1088             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
1089
1090             /* LENNARD-JONES DISPERSION/REPULSION */
1091
1092             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1093             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
1094             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1095             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
1096             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1097
1098             d                = _mm_sub_ps(r00,rswitch);
1099             d                = _mm_max_ps(d,_mm_setzero_ps());
1100             d2               = _mm_mul_ps(d,d);
1101             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)))))));
1102
1103             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1104
1105             /* Evaluate switch function */
1106             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1107             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1108             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1109
1110             fscal            = _mm_add_ps(felec,fvdw);
1111
1112             fscal            = _mm_and_ps(fscal,cutoff_mask);
1113
1114             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1115
1116             /* Calculate temporary vectorial force */
1117             tx               = _mm_mul_ps(fscal,dx00);
1118             ty               = _mm_mul_ps(fscal,dy00);
1119             tz               = _mm_mul_ps(fscal,dz00);
1120
1121             /* Update vectorial force */
1122             fix0             = _mm_add_ps(fix0,tx);
1123             fiy0             = _mm_add_ps(fiy0,ty);
1124             fiz0             = _mm_add_ps(fiz0,tz);
1125
1126             fjx0             = _mm_add_ps(fjx0,tx);
1127             fjy0             = _mm_add_ps(fjy0,ty);
1128             fjz0             = _mm_add_ps(fjz0,tz);
1129
1130             }
1131
1132             /**************************
1133              * CALCULATE INTERACTIONS *
1134              **************************/
1135
1136             if (gmx_mm_any_lt(rsq10,rcutoff2))
1137             {
1138
1139             /* Compute parameters for interactions between i and j atoms */
1140             qq10             = _mm_mul_ps(iq1,jq0);
1141
1142             /* REACTION-FIELD ELECTROSTATICS */
1143             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1144
1145             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1146
1147             fscal            = felec;
1148
1149             fscal            = _mm_and_ps(fscal,cutoff_mask);
1150
1151             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1152
1153             /* Calculate temporary vectorial force */
1154             tx               = _mm_mul_ps(fscal,dx10);
1155             ty               = _mm_mul_ps(fscal,dy10);
1156             tz               = _mm_mul_ps(fscal,dz10);
1157
1158             /* Update vectorial force */
1159             fix1             = _mm_add_ps(fix1,tx);
1160             fiy1             = _mm_add_ps(fiy1,ty);
1161             fiz1             = _mm_add_ps(fiz1,tz);
1162
1163             fjx0             = _mm_add_ps(fjx0,tx);
1164             fjy0             = _mm_add_ps(fjy0,ty);
1165             fjz0             = _mm_add_ps(fjz0,tz);
1166
1167             }
1168
1169             /**************************
1170              * CALCULATE INTERACTIONS *
1171              **************************/
1172
1173             if (gmx_mm_any_lt(rsq20,rcutoff2))
1174             {
1175
1176             /* Compute parameters for interactions between i and j atoms */
1177             qq20             = _mm_mul_ps(iq2,jq0);
1178
1179             /* REACTION-FIELD ELECTROSTATICS */
1180             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1181
1182             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1183
1184             fscal            = felec;
1185
1186             fscal            = _mm_and_ps(fscal,cutoff_mask);
1187
1188             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1189
1190             /* Calculate temporary vectorial force */
1191             tx               = _mm_mul_ps(fscal,dx20);
1192             ty               = _mm_mul_ps(fscal,dy20);
1193             tz               = _mm_mul_ps(fscal,dz20);
1194
1195             /* Update vectorial force */
1196             fix2             = _mm_add_ps(fix2,tx);
1197             fiy2             = _mm_add_ps(fiy2,ty);
1198             fiz2             = _mm_add_ps(fiz2,tz);
1199
1200             fjx0             = _mm_add_ps(fjx0,tx);
1201             fjy0             = _mm_add_ps(fjy0,ty);
1202             fjz0             = _mm_add_ps(fjz0,tz);
1203
1204             }
1205
1206             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1207             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1208             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1209             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1210
1211             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1212
1213             /* Inner loop uses 122 flops */
1214         }
1215
1216         /* End of innermost loop */
1217
1218         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1219                                               f+i_coord_offset,fshift+i_shift_offset);
1220
1221         /* Increment number of inner iterations */
1222         inneriter                  += j_index_end - j_index_start;
1223
1224         /* Outer loop uses 18 flops */
1225     }
1226
1227     /* Increment number of outer iterations */
1228     outeriter        += nri;
1229
1230     /* Update outer/inner flops */
1231
1232     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*122);
1233 }