src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_sse2_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse2_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_VF_sse2_double
  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_sse2_double
  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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwioffset1;
  82     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  83     int              vdwioffset2;
  84     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  85     int              vdwjidx0A,vdwjidx0B;
  86     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  89     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  90     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     int              nvdwtype;
  93     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  94     int              *vdwtype;
  95     real             *vdwparam;
  96     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  97     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  98     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  99     real             rswitch_scalar,d_scalar;
 100     __m128d          dummy_mask,cutoff_mask;
 101     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 102     __m128d          one     = _mm_set1_pd(1.0);
 103     __m128d          two     = _mm_set1_pd(2.0);
 104     x                = xx[0];
 105     f                = ff[0];
 106
 107     nri              = nlist->nri;
 108     iinr             = nlist->iinr;
 109     jindex           = nlist->jindex;
 110     jjnr             = nlist->jjnr;
 111     shiftidx         = nlist->shift;
 112     gid              = nlist->gid;
 113     shiftvec         = fr->shift_vec[0];
 114     fshift           = fr->fshift[0];
 115     facel            = _mm_set1_pd(fr->ic->epsfac);
 116     charge           = mdatoms->chargeA;
 117     krf              = _mm_set1_pd(fr->ic->k_rf);
 118     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 119     crf              = _mm_set1_pd(fr->ic->c_rf);
 120     nvdwtype         = fr->ntype;
 121     vdwparam         = fr->nbfp;
 122     vdwtype          = mdatoms->typeA;
 123
 124     /* Setup water-specific parameters */
 125     inr              = nlist->iinr[0];
 126     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 127     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 128     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 129     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 130
 131     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 132     rcutoff_scalar   = fr->ic->rcoulomb;
 133     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 134     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 135
 136     rswitch_scalar   = fr->ic->rvdw_switch;
 137     rswitch          = _mm_set1_pd(rswitch_scalar);
 138     /* Setup switch parameters */
 139     d_scalar         = rcutoff_scalar-rswitch_scalar;
 140     d                = _mm_set1_pd(d_scalar);
 141     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 142     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 143     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 144     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 145     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 146     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 147
 148     /* Avoid stupid compiler warnings */
 149     jnrA = jnrB = 0;
 150     j_coord_offsetA = 0;
 151     j_coord_offsetB = 0;
 152
 153     outeriter        = 0;
 154     inneriter        = 0;
 155
 156     /* Start outer loop over neighborlists */
 157     for(iidx=0; iidx<nri; iidx++)
 158     {
 159         /* Load shift vector for this list */
 160         i_shift_offset   = DIM*shiftidx[iidx];
 161
 162         /* Load limits for loop over neighbors */
 163         j_index_start    = jindex[iidx];
 164         j_index_end      = jindex[iidx+1];
 165
 166         /* Get outer coordinate index */
 167         inr              = iinr[iidx];
 168         i_coord_offset   = DIM*inr;
 169
 170         /* Load i particle coords and add shift vector */
 171         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 172                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 173
 174         fix0             = _mm_setzero_pd();
 175         fiy0             = _mm_setzero_pd();
 176         fiz0             = _mm_setzero_pd();
 177         fix1             = _mm_setzero_pd();
 178         fiy1             = _mm_setzero_pd();
 179         fiz1             = _mm_setzero_pd();
 180         fix2             = _mm_setzero_pd();
 181         fiy2             = _mm_setzero_pd();
 182         fiz2             = _mm_setzero_pd();
 183
 184         /* Reset potential sums */
 185         velecsum         = _mm_setzero_pd();
 186         vvdwsum          = _mm_setzero_pd();
 187
 188         /* Start inner kernel loop */
 189         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 190         {
 191
 192             /* Get j neighbor index, and coordinate index */
 193             jnrA             = jjnr[jidx];
 194             jnrB             = jjnr[jidx+1];
 195             j_coord_offsetA  = DIM*jnrA;
 196             j_coord_offsetB  = DIM*jnrB;
 197
 198             /* load j atom coordinates */
 199             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 200                                               &jx0,&jy0,&jz0);
 201
 202             /* Calculate displacement vector */
 203             dx00             = _mm_sub_pd(ix0,jx0);
 204             dy00             = _mm_sub_pd(iy0,jy0);
 205             dz00             = _mm_sub_pd(iz0,jz0);
 206             dx10             = _mm_sub_pd(ix1,jx0);
 207             dy10             = _mm_sub_pd(iy1,jy0);
 208             dz10             = _mm_sub_pd(iz1,jz0);
 209             dx20             = _mm_sub_pd(ix2,jx0);
 210             dy20             = _mm_sub_pd(iy2,jy0);
 211             dz20             = _mm_sub_pd(iz2,jz0);
 212
 213             /* Calculate squared distance and things based on it */
 214             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 215             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 216             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 217
 218             rinv00           = sse2_invsqrt_d(rsq00);
 219             rinv10           = sse2_invsqrt_d(rsq10);
 220             rinv20           = sse2_invsqrt_d(rsq20);
 221
 222             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 223             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 224             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 225
 226             /* Load parameters for j particles */
 227             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 228             vdwjidx0A        = 2*vdwtype[jnrA+0];
 229             vdwjidx0B        = 2*vdwtype[jnrB+0];
 230
 231             fjx0             = _mm_setzero_pd();
 232             fjy0             = _mm_setzero_pd();
 233             fjz0             = _mm_setzero_pd();
 234
 235             /**************************
 236              * CALCULATE INTERACTIONS *
 237              **************************/
 238
 239             if (gmx_mm_any_lt(rsq00,rcutoff2))
 240             {
 241
 242             r00              = _mm_mul_pd(rsq00,rinv00);
 243
 244             /* Compute parameters for interactions between i and j atoms */
 245             qq00             = _mm_mul_pd(iq0,jq0);
 246             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 247                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 248
 249             /* REACTION-FIELD ELECTROSTATICS */
 250             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 251             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 252
 253             /* LENNARD-JONES DISPERSION/REPULSION */
 254
 255             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 256             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 257             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 258             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 259             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 260
 261             d                = _mm_sub_pd(r00,rswitch);
 262             d                = _mm_max_pd(d,_mm_setzero_pd());
 263             d2               = _mm_mul_pd(d,d);
 264             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 265
 266             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 267
 268             /* Evaluate switch function */
 269             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 270             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 271             vvdw             = _mm_mul_pd(vvdw,sw);
 272             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 273
 274             /* Update potential sum for this i atom from the interaction with this j atom. */
 275             velec            = _mm_and_pd(velec,cutoff_mask);
 276             velecsum         = _mm_add_pd(velecsum,velec);
 277             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 278             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 279
 280             fscal            = _mm_add_pd(felec,fvdw);
 281
 282             fscal            = _mm_and_pd(fscal,cutoff_mask);
 283
 284             /* Calculate temporary vectorial force */
 285             tx               = _mm_mul_pd(fscal,dx00);
 286             ty               = _mm_mul_pd(fscal,dy00);
 287             tz               = _mm_mul_pd(fscal,dz00);
 288
 289             /* Update vectorial force */
 290             fix0             = _mm_add_pd(fix0,tx);
 291             fiy0             = _mm_add_pd(fiy0,ty);
 292             fiz0             = _mm_add_pd(fiz0,tz);
 293
 294             fjx0             = _mm_add_pd(fjx0,tx);
 295             fjy0             = _mm_add_pd(fjy0,ty);
 296             fjz0             = _mm_add_pd(fjz0,tz);
 297
 298             }
 299
 300             /**************************
 301              * CALCULATE INTERACTIONS *
 302              **************************/
 303
 304             if (gmx_mm_any_lt(rsq10,rcutoff2))
 305             {
 306
 307             /* Compute parameters for interactions between i and j atoms */
 308             qq10             = _mm_mul_pd(iq1,jq0);
 309
 310             /* REACTION-FIELD ELECTROSTATICS */
 311             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 312             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 313
 314             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 315
 316             /* Update potential sum for this i atom from the interaction with this j atom. */
 317             velec            = _mm_and_pd(velec,cutoff_mask);
 318             velecsum         = _mm_add_pd(velecsum,velec);
 319
 320             fscal            = felec;
 321
 322             fscal            = _mm_and_pd(fscal,cutoff_mask);
 323
 324             /* Calculate temporary vectorial force */
 325             tx               = _mm_mul_pd(fscal,dx10);
 326             ty               = _mm_mul_pd(fscal,dy10);
 327             tz               = _mm_mul_pd(fscal,dz10);
 328
 329             /* Update vectorial force */
 330             fix1             = _mm_add_pd(fix1,tx);
 331             fiy1             = _mm_add_pd(fiy1,ty);
 332             fiz1             = _mm_add_pd(fiz1,tz);
 333
 334             fjx0             = _mm_add_pd(fjx0,tx);
 335             fjy0             = _mm_add_pd(fjy0,ty);
 336             fjz0             = _mm_add_pd(fjz0,tz);
 337
 338             }
 339
 340             /**************************
 341              * CALCULATE INTERACTIONS *
 342              **************************/
 343
 344             if (gmx_mm_any_lt(rsq20,rcutoff2))
 345             {
 346
 347             /* Compute parameters for interactions between i and j atoms */
 348             qq20             = _mm_mul_pd(iq2,jq0);
 349
 350             /* REACTION-FIELD ELECTROSTATICS */
 351             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 352             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 353
 354             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 355
 356             /* Update potential sum for this i atom from the interaction with this j atom. */
 357             velec            = _mm_and_pd(velec,cutoff_mask);
 358             velecsum         = _mm_add_pd(velecsum,velec);
 359
 360             fscal            = felec;
 361
 362             fscal            = _mm_and_pd(fscal,cutoff_mask);
 363
 364             /* Calculate temporary vectorial force */
 365             tx               = _mm_mul_pd(fscal,dx20);
 366             ty               = _mm_mul_pd(fscal,dy20);
 367             tz               = _mm_mul_pd(fscal,dz20);
 368
 369             /* Update vectorial force */
 370             fix2             = _mm_add_pd(fix2,tx);
 371             fiy2             = _mm_add_pd(fiy2,ty);
 372             fiz2             = _mm_add_pd(fiz2,tz);
 373
 374             fjx0             = _mm_add_pd(fjx0,tx);
 375             fjy0             = _mm_add_pd(fjy0,ty);
 376             fjz0             = _mm_add_pd(fjz0,tz);
 377
 378             }
 379
 380             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 381
 382             /* Inner loop uses 145 flops */
 383         }
 384
 385         if(jidx<j_index_end)
 386         {
 387
 388             jnrA             = jjnr[jidx];
 389             j_coord_offsetA  = DIM*jnrA;
 390
 391             /* load j atom coordinates */
 392             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 393                                               &jx0,&jy0,&jz0);
 394
 395             /* Calculate displacement vector */
 396             dx00             = _mm_sub_pd(ix0,jx0);
 397             dy00             = _mm_sub_pd(iy0,jy0);
 398             dz00             = _mm_sub_pd(iz0,jz0);
 399             dx10             = _mm_sub_pd(ix1,jx0);
 400             dy10             = _mm_sub_pd(iy1,jy0);
 401             dz10             = _mm_sub_pd(iz1,jz0);
 402             dx20             = _mm_sub_pd(ix2,jx0);
 403             dy20             = _mm_sub_pd(iy2,jy0);
 404             dz20             = _mm_sub_pd(iz2,jz0);
 405
 406             /* Calculate squared distance and things based on it */
 407             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 408             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 409             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 410
 411             rinv00           = sse2_invsqrt_d(rsq00);
 412             rinv10           = sse2_invsqrt_d(rsq10);
 413             rinv20           = sse2_invsqrt_d(rsq20);
 414
 415             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 416             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 417             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 418
 419             /* Load parameters for j particles */
 420             jq0              = _mm_load_sd(charge+jnrA+0);
 421             vdwjidx0A        = 2*vdwtype[jnrA+0];
 422
 423             fjx0             = _mm_setzero_pd();
 424             fjy0             = _mm_setzero_pd();
 425             fjz0             = _mm_setzero_pd();
 426
 427             /**************************
 428              * CALCULATE INTERACTIONS *
 429              **************************/
 430
 431             if (gmx_mm_any_lt(rsq00,rcutoff2))
 432             {
 433
 434             r00              = _mm_mul_pd(rsq00,rinv00);
 435
 436             /* Compute parameters for interactions between i and j atoms */
 437             qq00             = _mm_mul_pd(iq0,jq0);
 438             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 439
 440             /* REACTION-FIELD ELECTROSTATICS */
 441             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 442             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 443
 444             /* LENNARD-JONES DISPERSION/REPULSION */
 445
 446             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 447             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 448             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 449             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 450             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 451
 452             d                = _mm_sub_pd(r00,rswitch);
 453             d                = _mm_max_pd(d,_mm_setzero_pd());
 454             d2               = _mm_mul_pd(d,d);
 455             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 456
 457             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 458
 459             /* Evaluate switch function */
 460             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 461             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 462             vvdw             = _mm_mul_pd(vvdw,sw);
 463             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 464
 465             /* Update potential sum for this i atom from the interaction with this j atom. */
 466             velec            = _mm_and_pd(velec,cutoff_mask);
 467             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 468             velecsum         = _mm_add_pd(velecsum,velec);
 469             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 470             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 471             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 472
 473             fscal            = _mm_add_pd(felec,fvdw);
 474
 475             fscal            = _mm_and_pd(fscal,cutoff_mask);
 476
 477             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 478
 479             /* Calculate temporary vectorial force */
 480             tx               = _mm_mul_pd(fscal,dx00);
 481             ty               = _mm_mul_pd(fscal,dy00);
 482             tz               = _mm_mul_pd(fscal,dz00);
 483
 484             /* Update vectorial force */
 485             fix0             = _mm_add_pd(fix0,tx);
 486             fiy0             = _mm_add_pd(fiy0,ty);
 487             fiz0             = _mm_add_pd(fiz0,tz);
 488
 489             fjx0             = _mm_add_pd(fjx0,tx);
 490             fjy0             = _mm_add_pd(fjy0,ty);
 491             fjz0             = _mm_add_pd(fjz0,tz);
 492
 493             }
 494
 495             /**************************
 496              * CALCULATE INTERACTIONS *
 497              **************************/
 498
 499             if (gmx_mm_any_lt(rsq10,rcutoff2))
 500             {
 501
 502             /* Compute parameters for interactions between i and j atoms */
 503             qq10             = _mm_mul_pd(iq1,jq0);
 504
 505             /* REACTION-FIELD ELECTROSTATICS */
 506             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 507             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 508
 509             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 510
 511             /* Update potential sum for this i atom from the interaction with this j atom. */
 512             velec            = _mm_and_pd(velec,cutoff_mask);
 513             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 514             velecsum         = _mm_add_pd(velecsum,velec);
 515
 516             fscal            = felec;
 517
 518             fscal            = _mm_and_pd(fscal,cutoff_mask);
 519
 520             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 521
 522             /* Calculate temporary vectorial force */
 523             tx               = _mm_mul_pd(fscal,dx10);
 524             ty               = _mm_mul_pd(fscal,dy10);
 525             tz               = _mm_mul_pd(fscal,dz10);
 526
 527             /* Update vectorial force */
 528             fix1             = _mm_add_pd(fix1,tx);
 529             fiy1             = _mm_add_pd(fiy1,ty);
 530             fiz1             = _mm_add_pd(fiz1,tz);
 531
 532             fjx0             = _mm_add_pd(fjx0,tx);
 533             fjy0             = _mm_add_pd(fjy0,ty);
 534             fjz0             = _mm_add_pd(fjz0,tz);
 535
 536             }
 537
 538             /**************************
 539              * CALCULATE INTERACTIONS *
 540              **************************/
 541
 542             if (gmx_mm_any_lt(rsq20,rcutoff2))
 543             {
 544
 545             /* Compute parameters for interactions between i and j atoms */
 546             qq20             = _mm_mul_pd(iq2,jq0);
 547
 548             /* REACTION-FIELD ELECTROSTATICS */
 549             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 550             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 551
 552             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 553
 554             /* Update potential sum for this i atom from the interaction with this j atom. */
 555             velec            = _mm_and_pd(velec,cutoff_mask);
 556             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 557             velecsum         = _mm_add_pd(velecsum,velec);
 558
 559             fscal            = felec;
 560
 561             fscal            = _mm_and_pd(fscal,cutoff_mask);
 562
 563             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 564
 565             /* Calculate temporary vectorial force */
 566             tx               = _mm_mul_pd(fscal,dx20);
 567             ty               = _mm_mul_pd(fscal,dy20);
 568             tz               = _mm_mul_pd(fscal,dz20);
 569
 570             /* Update vectorial force */
 571             fix2             = _mm_add_pd(fix2,tx);
 572             fiy2             = _mm_add_pd(fiy2,ty);
 573             fiz2             = _mm_add_pd(fiz2,tz);
 574
 575             fjx0             = _mm_add_pd(fjx0,tx);
 576             fjy0             = _mm_add_pd(fjy0,ty);
 577             fjz0             = _mm_add_pd(fjz0,tz);
 578
 579             }
 580
 581             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 582
 583             /* Inner loop uses 145 flops */
 584         }
 585
 586         /* End of innermost loop */
 587
 588         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 589                                               f+i_coord_offset,fshift+i_shift_offset);
 590
 591         ggid                        = gid[iidx];
 592         /* Update potential energies */
 593         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 594         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 595
 596         /* Increment number of inner iterations */
 597         inneriter                  += j_index_end - j_index_start;
 598
 599         /* Outer loop uses 20 flops */
 600     }
 601
 602     /* Increment number of outer iterations */
 603     outeriter        += nri;
 604
 605     /* Update outer/inner flops */
 606
 607     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*145);
 608 }
 609 /*
 610  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_sse2_double
 611  * Electrostatics interaction: ReactionField
 612  * VdW interaction:            LennardJones
 613  * Geometry:                   Water3-Particle
 614  * Calculate force/pot:        Force
 615  */
 616 void
 617 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_sse2_double
 618                     (t_nblist                    * gmx_restrict       nlist,
 619                      rvec                        * gmx_restrict          xx,
 620                      rvec                        * gmx_restrict          ff,
 621                      struct t_forcerec           * gmx_restrict          fr,
 622                      t_mdatoms                   * gmx_restrict     mdatoms,
 623                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 624                      t_nrnb                      * gmx_restrict        nrnb)
 625 {
 626     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 627      * just 0 for non-waters.
 628      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 629      * jnr indices corresponding to data put in the four positions in the SIMD register.
 630      */
 631     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 632     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 633     int              jnrA,jnrB;
 634     int              j_coord_offsetA,j_coord_offsetB;
 635     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 636     real             rcutoff_scalar;
 637     real             *shiftvec,*fshift,*x,*f;
 638     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 639     int              vdwioffset0;
 640     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 641     int              vdwioffset1;
 642     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 643     int              vdwioffset2;
 644     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 645     int              vdwjidx0A,vdwjidx0B;
 646     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 647     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 648     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 649     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 650     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 651     real             *charge;
 652     int              nvdwtype;
 653     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 654     int              *vdwtype;
 655     real             *vdwparam;
 656     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 657     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 658     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 659     real             rswitch_scalar,d_scalar;
 660     __m128d          dummy_mask,cutoff_mask;
 661     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 662     __m128d          one     = _mm_set1_pd(1.0);
 663     __m128d          two     = _mm_set1_pd(2.0);
 664     x                = xx[0];
 665     f                = ff[0];
 666
 667     nri              = nlist->nri;
 668     iinr             = nlist->iinr;
 669     jindex           = nlist->jindex;
 670     jjnr             = nlist->jjnr;
 671     shiftidx         = nlist->shift;
 672     gid              = nlist->gid;
 673     shiftvec         = fr->shift_vec[0];
 674     fshift           = fr->fshift[0];
 675     facel            = _mm_set1_pd(fr->ic->epsfac);
 676     charge           = mdatoms->chargeA;
 677     krf              = _mm_set1_pd(fr->ic->k_rf);
 678     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 679     crf              = _mm_set1_pd(fr->ic->c_rf);
 680     nvdwtype         = fr->ntype;
 681     vdwparam         = fr->nbfp;
 682     vdwtype          = mdatoms->typeA;
 683
 684     /* Setup water-specific parameters */
 685     inr              = nlist->iinr[0];
 686     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 687     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 688     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 689     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 690
 691     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 692     rcutoff_scalar   = fr->ic->rcoulomb;
 693     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 694     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 695
 696     rswitch_scalar   = fr->ic->rvdw_switch;
 697     rswitch          = _mm_set1_pd(rswitch_scalar);
 698     /* Setup switch parameters */
 699     d_scalar         = rcutoff_scalar-rswitch_scalar;
 700     d                = _mm_set1_pd(d_scalar);
 701     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 702     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 703     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 704     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 705     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 706     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 707
 708     /* Avoid stupid compiler warnings */
 709     jnrA = jnrB = 0;
 710     j_coord_offsetA = 0;
 711     j_coord_offsetB = 0;
 712
 713     outeriter        = 0;
 714     inneriter        = 0;
 715
 716     /* Start outer loop over neighborlists */
 717     for(iidx=0; iidx<nri; iidx++)
 718     {
 719         /* Load shift vector for this list */
 720         i_shift_offset   = DIM*shiftidx[iidx];
 721
 722         /* Load limits for loop over neighbors */
 723         j_index_start    = jindex[iidx];
 724         j_index_end      = jindex[iidx+1];
 725
 726         /* Get outer coordinate index */
 727         inr              = iinr[iidx];
 728         i_coord_offset   = DIM*inr;
 729
 730         /* Load i particle coords and add shift vector */
 731         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 732                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 733
 734         fix0             = _mm_setzero_pd();
 735         fiy0             = _mm_setzero_pd();
 736         fiz0             = _mm_setzero_pd();
 737         fix1             = _mm_setzero_pd();
 738         fiy1             = _mm_setzero_pd();
 739         fiz1             = _mm_setzero_pd();
 740         fix2             = _mm_setzero_pd();
 741         fiy2             = _mm_setzero_pd();
 742         fiz2             = _mm_setzero_pd();
 743
 744         /* Start inner kernel loop */
 745         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 746         {
 747
 748             /* Get j neighbor index, and coordinate index */
 749             jnrA             = jjnr[jidx];
 750             jnrB             = jjnr[jidx+1];
 751             j_coord_offsetA  = DIM*jnrA;
 752             j_coord_offsetB  = DIM*jnrB;
 753
 754             /* load j atom coordinates */
 755             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 756                                               &jx0,&jy0,&jz0);
 757
 758             /* Calculate displacement vector */
 759             dx00             = _mm_sub_pd(ix0,jx0);
 760             dy00             = _mm_sub_pd(iy0,jy0);
 761             dz00             = _mm_sub_pd(iz0,jz0);
 762             dx10             = _mm_sub_pd(ix1,jx0);
 763             dy10             = _mm_sub_pd(iy1,jy0);
 764             dz10             = _mm_sub_pd(iz1,jz0);
 765             dx20             = _mm_sub_pd(ix2,jx0);
 766             dy20             = _mm_sub_pd(iy2,jy0);
 767             dz20             = _mm_sub_pd(iz2,jz0);
 768
 769             /* Calculate squared distance and things based on it */
 770             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 771             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 772             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 773
 774             rinv00           = sse2_invsqrt_d(rsq00);
 775             rinv10           = sse2_invsqrt_d(rsq10);
 776             rinv20           = sse2_invsqrt_d(rsq20);
 777
 778             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 779             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 780             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 781
 782             /* Load parameters for j particles */
 783             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 784             vdwjidx0A        = 2*vdwtype[jnrA+0];
 785             vdwjidx0B        = 2*vdwtype[jnrB+0];
 786
 787             fjx0             = _mm_setzero_pd();
 788             fjy0             = _mm_setzero_pd();
 789             fjz0             = _mm_setzero_pd();
 790
 791             /**************************
 792              * CALCULATE INTERACTIONS *
 793              **************************/
 794
 795             if (gmx_mm_any_lt(rsq00,rcutoff2))
 796             {
 797
 798             r00              = _mm_mul_pd(rsq00,rinv00);
 799
 800             /* Compute parameters for interactions between i and j atoms */
 801             qq00             = _mm_mul_pd(iq0,jq0);
 802             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 803                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 804
 805             /* REACTION-FIELD ELECTROSTATICS */
 806             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 807
 808             /* LENNARD-JONES DISPERSION/REPULSION */
 809
 810             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 811             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 812             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 813             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 814             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 815
 816             d                = _mm_sub_pd(r00,rswitch);
 817             d                = _mm_max_pd(d,_mm_setzero_pd());
 818             d2               = _mm_mul_pd(d,d);
 819             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 820
 821             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 822
 823             /* Evaluate switch function */
 824             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 825             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 826             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 827
 828             fscal            = _mm_add_pd(felec,fvdw);
 829
 830             fscal            = _mm_and_pd(fscal,cutoff_mask);
 831
 832             /* Calculate temporary vectorial force */
 833             tx               = _mm_mul_pd(fscal,dx00);
 834             ty               = _mm_mul_pd(fscal,dy00);
 835             tz               = _mm_mul_pd(fscal,dz00);
 836
 837             /* Update vectorial force */
 838             fix0             = _mm_add_pd(fix0,tx);
 839             fiy0             = _mm_add_pd(fiy0,ty);
 840             fiz0             = _mm_add_pd(fiz0,tz);
 841
 842             fjx0             = _mm_add_pd(fjx0,tx);
 843             fjy0             = _mm_add_pd(fjy0,ty);
 844             fjz0             = _mm_add_pd(fjz0,tz);
 845
 846             }
 847
 848             /**************************
 849              * CALCULATE INTERACTIONS *
 850              **************************/
 851
 852             if (gmx_mm_any_lt(rsq10,rcutoff2))
 853             {
 854
 855             /* Compute parameters for interactions between i and j atoms */
 856             qq10             = _mm_mul_pd(iq1,jq0);
 857
 858             /* REACTION-FIELD ELECTROSTATICS */
 859             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 860
 861             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 862
 863             fscal            = felec;
 864
 865             fscal            = _mm_and_pd(fscal,cutoff_mask);
 866
 867             /* Calculate temporary vectorial force */
 868             tx               = _mm_mul_pd(fscal,dx10);
 869             ty               = _mm_mul_pd(fscal,dy10);
 870             tz               = _mm_mul_pd(fscal,dz10);
 871
 872             /* Update vectorial force */
 873             fix1             = _mm_add_pd(fix1,tx);
 874             fiy1             = _mm_add_pd(fiy1,ty);
 875             fiz1             = _mm_add_pd(fiz1,tz);
 876
 877             fjx0             = _mm_add_pd(fjx0,tx);
 878             fjy0             = _mm_add_pd(fjy0,ty);
 879             fjz0             = _mm_add_pd(fjz0,tz);
 880
 881             }
 882
 883             /**************************
 884              * CALCULATE INTERACTIONS *
 885              **************************/
 886
 887             if (gmx_mm_any_lt(rsq20,rcutoff2))
 888             {
 889
 890             /* Compute parameters for interactions between i and j atoms */
 891             qq20             = _mm_mul_pd(iq2,jq0);
 892
 893             /* REACTION-FIELD ELECTROSTATICS */
 894             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 895
 896             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 897
 898             fscal            = felec;
 899
 900             fscal            = _mm_and_pd(fscal,cutoff_mask);
 901
 902             /* Calculate temporary vectorial force */
 903             tx               = _mm_mul_pd(fscal,dx20);
 904             ty               = _mm_mul_pd(fscal,dy20);
 905             tz               = _mm_mul_pd(fscal,dz20);
 906
 907             /* Update vectorial force */
 908             fix2             = _mm_add_pd(fix2,tx);
 909             fiy2             = _mm_add_pd(fiy2,ty);
 910             fiz2             = _mm_add_pd(fiz2,tz);
 911
 912             fjx0             = _mm_add_pd(fjx0,tx);
 913             fjy0             = _mm_add_pd(fjy0,ty);
 914             fjz0             = _mm_add_pd(fjz0,tz);
 915
 916             }
 917
 918             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 919
 920             /* Inner loop uses 124 flops */
 921         }
 922
 923         if(jidx<j_index_end)
 924         {
 925
 926             jnrA             = jjnr[jidx];
 927             j_coord_offsetA  = DIM*jnrA;
 928
 929             /* load j atom coordinates */
 930             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 931                                               &jx0,&jy0,&jz0);
 932
 933             /* Calculate displacement vector */
 934             dx00             = _mm_sub_pd(ix0,jx0);
 935             dy00             = _mm_sub_pd(iy0,jy0);
 936             dz00             = _mm_sub_pd(iz0,jz0);
 937             dx10             = _mm_sub_pd(ix1,jx0);
 938             dy10             = _mm_sub_pd(iy1,jy0);
 939             dz10             = _mm_sub_pd(iz1,jz0);
 940             dx20             = _mm_sub_pd(ix2,jx0);
 941             dy20             = _mm_sub_pd(iy2,jy0);
 942             dz20             = _mm_sub_pd(iz2,jz0);
 943
 944             /* Calculate squared distance and things based on it */
 945             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 946             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 947             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 948
 949             rinv00           = sse2_invsqrt_d(rsq00);
 950             rinv10           = sse2_invsqrt_d(rsq10);
 951             rinv20           = sse2_invsqrt_d(rsq20);
 952
 953             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 954             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 955             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 956
 957             /* Load parameters for j particles */
 958             jq0              = _mm_load_sd(charge+jnrA+0);
 959             vdwjidx0A        = 2*vdwtype[jnrA+0];
 960
 961             fjx0             = _mm_setzero_pd();
 962             fjy0             = _mm_setzero_pd();
 963             fjz0             = _mm_setzero_pd();
 964
 965             /**************************
 966              * CALCULATE INTERACTIONS *
 967              **************************/
 968
 969             if (gmx_mm_any_lt(rsq00,rcutoff2))
 970             {
 971
 972             r00              = _mm_mul_pd(rsq00,rinv00);
 973
 974             /* Compute parameters for interactions between i and j atoms */
 975             qq00             = _mm_mul_pd(iq0,jq0);
 976             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 977
 978             /* REACTION-FIELD ELECTROSTATICS */
 979             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 980
 981             /* LENNARD-JONES DISPERSION/REPULSION */
 982
 983             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 984             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 985             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 986             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 987             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 988
 989             d                = _mm_sub_pd(r00,rswitch);
 990             d                = _mm_max_pd(d,_mm_setzero_pd());
 991             d2               = _mm_mul_pd(d,d);
 992             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 993
 994             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 995
 996             /* Evaluate switch function */
 997             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 998             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 999             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1000
1001             fscal            = _mm_add_pd(felec,fvdw);
1002
1003             fscal            = _mm_and_pd(fscal,cutoff_mask);
1004
1005             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1006
1007             /* Calculate temporary vectorial force */
1008             tx               = _mm_mul_pd(fscal,dx00);
1009             ty               = _mm_mul_pd(fscal,dy00);
1010             tz               = _mm_mul_pd(fscal,dz00);
1011
1012             /* Update vectorial force */
1013             fix0             = _mm_add_pd(fix0,tx);
1014             fiy0             = _mm_add_pd(fiy0,ty);
1015             fiz0             = _mm_add_pd(fiz0,tz);
1016
1017             fjx0             = _mm_add_pd(fjx0,tx);
1018             fjy0             = _mm_add_pd(fjy0,ty);
1019             fjz0             = _mm_add_pd(fjz0,tz);
1020
1021             }
1022
1023             /**************************
1024              * CALCULATE INTERACTIONS *
1025              **************************/
1026
1027             if (gmx_mm_any_lt(rsq10,rcutoff2))
1028             {
1029
1030             /* Compute parameters for interactions between i and j atoms */
1031             qq10             = _mm_mul_pd(iq1,jq0);
1032
1033             /* REACTION-FIELD ELECTROSTATICS */
1034             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
1035
1036             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1037
1038             fscal            = felec;
1039
1040             fscal            = _mm_and_pd(fscal,cutoff_mask);
1041
1042             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1043
1044             /* Calculate temporary vectorial force */
1045             tx               = _mm_mul_pd(fscal,dx10);
1046             ty               = _mm_mul_pd(fscal,dy10);
1047             tz               = _mm_mul_pd(fscal,dz10);
1048
1049             /* Update vectorial force */
1050             fix1             = _mm_add_pd(fix1,tx);
1051             fiy1             = _mm_add_pd(fiy1,ty);
1052             fiz1             = _mm_add_pd(fiz1,tz);
1053
1054             fjx0             = _mm_add_pd(fjx0,tx);
1055             fjy0             = _mm_add_pd(fjy0,ty);
1056             fjz0             = _mm_add_pd(fjz0,tz);
1057
1058             }
1059
1060             /**************************
1061              * CALCULATE INTERACTIONS *
1062              **************************/
1063
1064             if (gmx_mm_any_lt(rsq20,rcutoff2))
1065             {
1066
1067             /* Compute parameters for interactions between i and j atoms */
1068             qq20             = _mm_mul_pd(iq2,jq0);
1069
1070             /* REACTION-FIELD ELECTROSTATICS */
1071             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
1072
1073             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1074
1075             fscal            = felec;
1076
1077             fscal            = _mm_and_pd(fscal,cutoff_mask);
1078
1079             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1080
1081             /* Calculate temporary vectorial force */
1082             tx               = _mm_mul_pd(fscal,dx20);
1083             ty               = _mm_mul_pd(fscal,dy20);
1084             tz               = _mm_mul_pd(fscal,dz20);
1085
1086             /* Update vectorial force */
1087             fix2             = _mm_add_pd(fix2,tx);
1088             fiy2             = _mm_add_pd(fiy2,ty);
1089             fiz2             = _mm_add_pd(fiz2,tz);
1090
1091             fjx0             = _mm_add_pd(fjx0,tx);
1092             fjy0             = _mm_add_pd(fjy0,ty);
1093             fjz0             = _mm_add_pd(fjz0,tz);
1094
1095             }
1096
1097             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1098
1099             /* Inner loop uses 124 flops */
1100         }
1101
1102         /* End of innermost loop */
1103
1104         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1105                                               f+i_coord_offset,fshift+i_shift_offset);
1106
1107         /* Increment number of inner iterations */
1108         inneriter                  += j_index_end - j_index_start;
1109
1110         /* Outer loop uses 18 flops */
1111     }
1112
1113     /* Increment number of outer iterations */
1114     outeriter        += nri;
1115
1116     /* Update outer/inner flops */
1117
1118     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*124);
1119 }