src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_sse4_1_double.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_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_sse4_1_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_VF_sse4_1_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_VdwLJSh_GeomW3P1_VF_sse4_1_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          dummy_mask,cutoff_mask;
  99     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 100     __m128d          one     = _mm_set1_pd(1.0);
 101     __m128d          two     = _mm_set1_pd(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = _mm_set1_pd(fr->ic->epsfac);
 114     charge           = mdatoms->chargeA;
 115     krf              = _mm_set1_pd(fr->ic->k_rf);
 116     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 117     crf              = _mm_set1_pd(fr->ic->c_rf);
 118     nvdwtype         = fr->ntype;
 119     vdwparam         = fr->nbfp;
 120     vdwtype          = mdatoms->typeA;
 121
 122     /* Setup water-specific parameters */
 123     inr              = nlist->iinr[0];
 124     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 125     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 126     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 127     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 128
 129     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 130     rcutoff_scalar   = fr->ic->rcoulomb;
 131     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 132     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 133
 134     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 135     rvdw             = _mm_set1_pd(fr->ic->rvdw);
 136
 137     /* Avoid stupid compiler warnings */
 138     jnrA = jnrB = 0;
 139     j_coord_offsetA = 0;
 140     j_coord_offsetB = 0;
 141
 142     outeriter        = 0;
 143     inneriter        = 0;
 144
 145     /* Start outer loop over neighborlists */
 146     for(iidx=0; iidx<nri; iidx++)
 147     {
 148         /* Load shift vector for this list */
 149         i_shift_offset   = DIM*shiftidx[iidx];
 150
 151         /* Load limits for loop over neighbors */
 152         j_index_start    = jindex[iidx];
 153         j_index_end      = jindex[iidx+1];
 154
 155         /* Get outer coordinate index */
 156         inr              = iinr[iidx];
 157         i_coord_offset   = DIM*inr;
 158
 159         /* Load i particle coords and add shift vector */
 160         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 161                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 162
 163         fix0             = _mm_setzero_pd();
 164         fiy0             = _mm_setzero_pd();
 165         fiz0             = _mm_setzero_pd();
 166         fix1             = _mm_setzero_pd();
 167         fiy1             = _mm_setzero_pd();
 168         fiz1             = _mm_setzero_pd();
 169         fix2             = _mm_setzero_pd();
 170         fiy2             = _mm_setzero_pd();
 171         fiz2             = _mm_setzero_pd();
 172
 173         /* Reset potential sums */
 174         velecsum         = _mm_setzero_pd();
 175         vvdwsum          = _mm_setzero_pd();
 176
 177         /* Start inner kernel loop */
 178         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 179         {
 180
 181             /* Get j neighbor index, and coordinate index */
 182             jnrA             = jjnr[jidx];
 183             jnrB             = jjnr[jidx+1];
 184             j_coord_offsetA  = DIM*jnrA;
 185             j_coord_offsetB  = DIM*jnrB;
 186
 187             /* load j atom coordinates */
 188             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 189                                               &jx0,&jy0,&jz0);
 190
 191             /* Calculate displacement vector */
 192             dx00             = _mm_sub_pd(ix0,jx0);
 193             dy00             = _mm_sub_pd(iy0,jy0);
 194             dz00             = _mm_sub_pd(iz0,jz0);
 195             dx10             = _mm_sub_pd(ix1,jx0);
 196             dy10             = _mm_sub_pd(iy1,jy0);
 197             dz10             = _mm_sub_pd(iz1,jz0);
 198             dx20             = _mm_sub_pd(ix2,jx0);
 199             dy20             = _mm_sub_pd(iy2,jy0);
 200             dz20             = _mm_sub_pd(iz2,jz0);
 201
 202             /* Calculate squared distance and things based on it */
 203             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 204             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 205             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 206
 207             rinv00           = sse41_invsqrt_d(rsq00);
 208             rinv10           = sse41_invsqrt_d(rsq10);
 209             rinv20           = sse41_invsqrt_d(rsq20);
 210
 211             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 212             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 213             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 214
 215             /* Load parameters for j particles */
 216             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 217             vdwjidx0A        = 2*vdwtype[jnrA+0];
 218             vdwjidx0B        = 2*vdwtype[jnrB+0];
 219
 220             fjx0             = _mm_setzero_pd();
 221             fjy0             = _mm_setzero_pd();
 222             fjz0             = _mm_setzero_pd();
 223
 224             /**************************
 225              * CALCULATE INTERACTIONS *
 226              **************************/
 227
 228             if (gmx_mm_any_lt(rsq00,rcutoff2))
 229             {
 230
 231             /* Compute parameters for interactions between i and j atoms */
 232             qq00             = _mm_mul_pd(iq0,jq0);
 233             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 234                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 235
 236             /* REACTION-FIELD ELECTROSTATICS */
 237             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 238             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 239
 240             /* LENNARD-JONES DISPERSION/REPULSION */
 241
 242             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 243             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 244             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 245             vvdw             = _mm_sub_pd(_mm_mul_pd( _mm_sub_pd(vvdw12 , _mm_mul_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 246                                           _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 247             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 248
 249             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 250
 251             /* Update potential sum for this i atom from the interaction with this j atom. */
 252             velec            = _mm_and_pd(velec,cutoff_mask);
 253             velecsum         = _mm_add_pd(velecsum,velec);
 254             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 255             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 256
 257             fscal            = _mm_add_pd(felec,fvdw);
 258
 259             fscal            = _mm_and_pd(fscal,cutoff_mask);
 260
 261             /* Calculate temporary vectorial force */
 262             tx               = _mm_mul_pd(fscal,dx00);
 263             ty               = _mm_mul_pd(fscal,dy00);
 264             tz               = _mm_mul_pd(fscal,dz00);
 265
 266             /* Update vectorial force */
 267             fix0             = _mm_add_pd(fix0,tx);
 268             fiy0             = _mm_add_pd(fiy0,ty);
 269             fiz0             = _mm_add_pd(fiz0,tz);
 270
 271             fjx0             = _mm_add_pd(fjx0,tx);
 272             fjy0             = _mm_add_pd(fjy0,ty);
 273             fjz0             = _mm_add_pd(fjz0,tz);
 274
 275             }
 276
 277             /**************************
 278              * CALCULATE INTERACTIONS *
 279              **************************/
 280
 281             if (gmx_mm_any_lt(rsq10,rcutoff2))
 282             {
 283
 284             /* Compute parameters for interactions between i and j atoms */
 285             qq10             = _mm_mul_pd(iq1,jq0);
 286
 287             /* REACTION-FIELD ELECTROSTATICS */
 288             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 289             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 290
 291             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 292
 293             /* Update potential sum for this i atom from the interaction with this j atom. */
 294             velec            = _mm_and_pd(velec,cutoff_mask);
 295             velecsum         = _mm_add_pd(velecsum,velec);
 296
 297             fscal            = felec;
 298
 299             fscal            = _mm_and_pd(fscal,cutoff_mask);
 300
 301             /* Calculate temporary vectorial force */
 302             tx               = _mm_mul_pd(fscal,dx10);
 303             ty               = _mm_mul_pd(fscal,dy10);
 304             tz               = _mm_mul_pd(fscal,dz10);
 305
 306             /* Update vectorial force */
 307             fix1             = _mm_add_pd(fix1,tx);
 308             fiy1             = _mm_add_pd(fiy1,ty);
 309             fiz1             = _mm_add_pd(fiz1,tz);
 310
 311             fjx0             = _mm_add_pd(fjx0,tx);
 312             fjy0             = _mm_add_pd(fjy0,ty);
 313             fjz0             = _mm_add_pd(fjz0,tz);
 314
 315             }
 316
 317             /**************************
 318              * CALCULATE INTERACTIONS *
 319              **************************/
 320
 321             if (gmx_mm_any_lt(rsq20,rcutoff2))
 322             {
 323
 324             /* Compute parameters for interactions between i and j atoms */
 325             qq20             = _mm_mul_pd(iq2,jq0);
 326
 327             /* REACTION-FIELD ELECTROSTATICS */
 328             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 329             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 330
 331             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 332
 333             /* Update potential sum for this i atom from the interaction with this j atom. */
 334             velec            = _mm_and_pd(velec,cutoff_mask);
 335             velecsum         = _mm_add_pd(velecsum,velec);
 336
 337             fscal            = felec;
 338
 339             fscal            = _mm_and_pd(fscal,cutoff_mask);
 340
 341             /* Calculate temporary vectorial force */
 342             tx               = _mm_mul_pd(fscal,dx20);
 343             ty               = _mm_mul_pd(fscal,dy20);
 344             tz               = _mm_mul_pd(fscal,dz20);
 345
 346             /* Update vectorial force */
 347             fix2             = _mm_add_pd(fix2,tx);
 348             fiy2             = _mm_add_pd(fiy2,ty);
 349             fiz2             = _mm_add_pd(fiz2,tz);
 350
 351             fjx0             = _mm_add_pd(fjx0,tx);
 352             fjy0             = _mm_add_pd(fjy0,ty);
 353             fjz0             = _mm_add_pd(fjz0,tz);
 354
 355             }
 356
 357             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 358
 359             /* Inner loop uses 129 flops */
 360         }
 361
 362         if(jidx<j_index_end)
 363         {
 364
 365             jnrA             = jjnr[jidx];
 366             j_coord_offsetA  = DIM*jnrA;
 367
 368             /* load j atom coordinates */
 369             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 370                                               &jx0,&jy0,&jz0);
 371
 372             /* Calculate displacement vector */
 373             dx00             = _mm_sub_pd(ix0,jx0);
 374             dy00             = _mm_sub_pd(iy0,jy0);
 375             dz00             = _mm_sub_pd(iz0,jz0);
 376             dx10             = _mm_sub_pd(ix1,jx0);
 377             dy10             = _mm_sub_pd(iy1,jy0);
 378             dz10             = _mm_sub_pd(iz1,jz0);
 379             dx20             = _mm_sub_pd(ix2,jx0);
 380             dy20             = _mm_sub_pd(iy2,jy0);
 381             dz20             = _mm_sub_pd(iz2,jz0);
 382
 383             /* Calculate squared distance and things based on it */
 384             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 385             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 386             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 387
 388             rinv00           = sse41_invsqrt_d(rsq00);
 389             rinv10           = sse41_invsqrt_d(rsq10);
 390             rinv20           = sse41_invsqrt_d(rsq20);
 391
 392             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 393             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 394             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 395
 396             /* Load parameters for j particles */
 397             jq0              = _mm_load_sd(charge+jnrA+0);
 398             vdwjidx0A        = 2*vdwtype[jnrA+0];
 399
 400             fjx0             = _mm_setzero_pd();
 401             fjy0             = _mm_setzero_pd();
 402             fjz0             = _mm_setzero_pd();
 403
 404             /**************************
 405              * CALCULATE INTERACTIONS *
 406              **************************/
 407
 408             if (gmx_mm_any_lt(rsq00,rcutoff2))
 409             {
 410
 411             /* Compute parameters for interactions between i and j atoms */
 412             qq00             = _mm_mul_pd(iq0,jq0);
 413             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 414
 415             /* REACTION-FIELD ELECTROSTATICS */
 416             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 417             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 418
 419             /* LENNARD-JONES DISPERSION/REPULSION */
 420
 421             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 422             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 423             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 424             vvdw             = _mm_sub_pd(_mm_mul_pd( _mm_sub_pd(vvdw12 , _mm_mul_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 425                                           _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 426             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 427
 428             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 429
 430             /* Update potential sum for this i atom from the interaction with this j atom. */
 431             velec            = _mm_and_pd(velec,cutoff_mask);
 432             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 433             velecsum         = _mm_add_pd(velecsum,velec);
 434             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 435             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 436             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 437
 438             fscal            = _mm_add_pd(felec,fvdw);
 439
 440             fscal            = _mm_and_pd(fscal,cutoff_mask);
 441
 442             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 443
 444             /* Calculate temporary vectorial force */
 445             tx               = _mm_mul_pd(fscal,dx00);
 446             ty               = _mm_mul_pd(fscal,dy00);
 447             tz               = _mm_mul_pd(fscal,dz00);
 448
 449             /* Update vectorial force */
 450             fix0             = _mm_add_pd(fix0,tx);
 451             fiy0             = _mm_add_pd(fiy0,ty);
 452             fiz0             = _mm_add_pd(fiz0,tz);
 453
 454             fjx0             = _mm_add_pd(fjx0,tx);
 455             fjy0             = _mm_add_pd(fjy0,ty);
 456             fjz0             = _mm_add_pd(fjz0,tz);
 457
 458             }
 459
 460             /**************************
 461              * CALCULATE INTERACTIONS *
 462              **************************/
 463
 464             if (gmx_mm_any_lt(rsq10,rcutoff2))
 465             {
 466
 467             /* Compute parameters for interactions between i and j atoms */
 468             qq10             = _mm_mul_pd(iq1,jq0);
 469
 470             /* REACTION-FIELD ELECTROSTATICS */
 471             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 472             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 473
 474             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 475
 476             /* Update potential sum for this i atom from the interaction with this j atom. */
 477             velec            = _mm_and_pd(velec,cutoff_mask);
 478             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 479             velecsum         = _mm_add_pd(velecsum,velec);
 480
 481             fscal            = felec;
 482
 483             fscal            = _mm_and_pd(fscal,cutoff_mask);
 484
 485             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 486
 487             /* Calculate temporary vectorial force */
 488             tx               = _mm_mul_pd(fscal,dx10);
 489             ty               = _mm_mul_pd(fscal,dy10);
 490             tz               = _mm_mul_pd(fscal,dz10);
 491
 492             /* Update vectorial force */
 493             fix1             = _mm_add_pd(fix1,tx);
 494             fiy1             = _mm_add_pd(fiy1,ty);
 495             fiz1             = _mm_add_pd(fiz1,tz);
 496
 497             fjx0             = _mm_add_pd(fjx0,tx);
 498             fjy0             = _mm_add_pd(fjy0,ty);
 499             fjz0             = _mm_add_pd(fjz0,tz);
 500
 501             }
 502
 503             /**************************
 504              * CALCULATE INTERACTIONS *
 505              **************************/
 506
 507             if (gmx_mm_any_lt(rsq20,rcutoff2))
 508             {
 509
 510             /* Compute parameters for interactions between i and j atoms */
 511             qq20             = _mm_mul_pd(iq2,jq0);
 512
 513             /* REACTION-FIELD ELECTROSTATICS */
 514             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 515             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 516
 517             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 518
 519             /* Update potential sum for this i atom from the interaction with this j atom. */
 520             velec            = _mm_and_pd(velec,cutoff_mask);
 521             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 522             velecsum         = _mm_add_pd(velecsum,velec);
 523
 524             fscal            = felec;
 525
 526             fscal            = _mm_and_pd(fscal,cutoff_mask);
 527
 528             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 529
 530             /* Calculate temporary vectorial force */
 531             tx               = _mm_mul_pd(fscal,dx20);
 532             ty               = _mm_mul_pd(fscal,dy20);
 533             tz               = _mm_mul_pd(fscal,dz20);
 534
 535             /* Update vectorial force */
 536             fix2             = _mm_add_pd(fix2,tx);
 537             fiy2             = _mm_add_pd(fiy2,ty);
 538             fiz2             = _mm_add_pd(fiz2,tz);
 539
 540             fjx0             = _mm_add_pd(fjx0,tx);
 541             fjy0             = _mm_add_pd(fjy0,ty);
 542             fjz0             = _mm_add_pd(fjz0,tz);
 543
 544             }
 545
 546             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 547
 548             /* Inner loop uses 129 flops */
 549         }
 550
 551         /* End of innermost loop */
 552
 553         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 554                                               f+i_coord_offset,fshift+i_shift_offset);
 555
 556         ggid                        = gid[iidx];
 557         /* Update potential energies */
 558         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 559         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 560
 561         /* Increment number of inner iterations */
 562         inneriter                  += j_index_end - j_index_start;
 563
 564         /* Outer loop uses 20 flops */
 565     }
 566
 567     /* Increment number of outer iterations */
 568     outeriter        += nri;
 569
 570     /* Update outer/inner flops */
 571
 572     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*129);
 573 }
 574 /*
 575  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_sse4_1_double
 576  * Electrostatics interaction: ReactionField
 577  * VdW interaction:            LennardJones
 578  * Geometry:                   Water3-Particle
 579  * Calculate force/pot:        Force
 580  */
 581 void
 582 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_sse4_1_double
 583                     (t_nblist                    * gmx_restrict       nlist,
 584                      rvec                        * gmx_restrict          xx,
 585                      rvec                        * gmx_restrict          ff,
 586                      struct t_forcerec           * gmx_restrict          fr,
 587                      t_mdatoms                   * gmx_restrict     mdatoms,
 588                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 589                      t_nrnb                      * gmx_restrict        nrnb)
 590 {
 591     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 592      * just 0 for non-waters.
 593      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 594      * jnr indices corresponding to data put in the four positions in the SIMD register.
 595      */
 596     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 597     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 598     int              jnrA,jnrB;
 599     int              j_coord_offsetA,j_coord_offsetB;
 600     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 601     real             rcutoff_scalar;
 602     real             *shiftvec,*fshift,*x,*f;
 603     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 604     int              vdwioffset0;
 605     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 606     int              vdwioffset1;
 607     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 608     int              vdwioffset2;
 609     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 610     int              vdwjidx0A,vdwjidx0B;
 611     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 612     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 613     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 614     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 615     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 616     real             *charge;
 617     int              nvdwtype;
 618     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 619     int              *vdwtype;
 620     real             *vdwparam;
 621     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 622     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 623     __m128d          dummy_mask,cutoff_mask;
 624     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 625     __m128d          one     = _mm_set1_pd(1.0);
 626     __m128d          two     = _mm_set1_pd(2.0);
 627     x                = xx[0];
 628     f                = ff[0];
 629
 630     nri              = nlist->nri;
 631     iinr             = nlist->iinr;
 632     jindex           = nlist->jindex;
 633     jjnr             = nlist->jjnr;
 634     shiftidx         = nlist->shift;
 635     gid              = nlist->gid;
 636     shiftvec         = fr->shift_vec[0];
 637     fshift           = fr->fshift[0];
 638     facel            = _mm_set1_pd(fr->ic->epsfac);
 639     charge           = mdatoms->chargeA;
 640     krf              = _mm_set1_pd(fr->ic->k_rf);
 641     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 642     crf              = _mm_set1_pd(fr->ic->c_rf);
 643     nvdwtype         = fr->ntype;
 644     vdwparam         = fr->nbfp;
 645     vdwtype          = mdatoms->typeA;
 646
 647     /* Setup water-specific parameters */
 648     inr              = nlist->iinr[0];
 649     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 650     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 651     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 652     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 653
 654     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 655     rcutoff_scalar   = fr->ic->rcoulomb;
 656     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 657     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 658
 659     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 660     rvdw             = _mm_set1_pd(fr->ic->rvdw);
 661
 662     /* Avoid stupid compiler warnings */
 663     jnrA = jnrB = 0;
 664     j_coord_offsetA = 0;
 665     j_coord_offsetB = 0;
 666
 667     outeriter        = 0;
 668     inneriter        = 0;
 669
 670     /* Start outer loop over neighborlists */
 671     for(iidx=0; iidx<nri; iidx++)
 672     {
 673         /* Load shift vector for this list */
 674         i_shift_offset   = DIM*shiftidx[iidx];
 675
 676         /* Load limits for loop over neighbors */
 677         j_index_start    = jindex[iidx];
 678         j_index_end      = jindex[iidx+1];
 679
 680         /* Get outer coordinate index */
 681         inr              = iinr[iidx];
 682         i_coord_offset   = DIM*inr;
 683
 684         /* Load i particle coords and add shift vector */
 685         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 686                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 687
 688         fix0             = _mm_setzero_pd();
 689         fiy0             = _mm_setzero_pd();
 690         fiz0             = _mm_setzero_pd();
 691         fix1             = _mm_setzero_pd();
 692         fiy1             = _mm_setzero_pd();
 693         fiz1             = _mm_setzero_pd();
 694         fix2             = _mm_setzero_pd();
 695         fiy2             = _mm_setzero_pd();
 696         fiz2             = _mm_setzero_pd();
 697
 698         /* Start inner kernel loop */
 699         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 700         {
 701
 702             /* Get j neighbor index, and coordinate index */
 703             jnrA             = jjnr[jidx];
 704             jnrB             = jjnr[jidx+1];
 705             j_coord_offsetA  = DIM*jnrA;
 706             j_coord_offsetB  = DIM*jnrB;
 707
 708             /* load j atom coordinates */
 709             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 710                                               &jx0,&jy0,&jz0);
 711
 712             /* Calculate displacement vector */
 713             dx00             = _mm_sub_pd(ix0,jx0);
 714             dy00             = _mm_sub_pd(iy0,jy0);
 715             dz00             = _mm_sub_pd(iz0,jz0);
 716             dx10             = _mm_sub_pd(ix1,jx0);
 717             dy10             = _mm_sub_pd(iy1,jy0);
 718             dz10             = _mm_sub_pd(iz1,jz0);
 719             dx20             = _mm_sub_pd(ix2,jx0);
 720             dy20             = _mm_sub_pd(iy2,jy0);
 721             dz20             = _mm_sub_pd(iz2,jz0);
 722
 723             /* Calculate squared distance and things based on it */
 724             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 725             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 726             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 727
 728             rinv00           = sse41_invsqrt_d(rsq00);
 729             rinv10           = sse41_invsqrt_d(rsq10);
 730             rinv20           = sse41_invsqrt_d(rsq20);
 731
 732             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 733             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 734             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 735
 736             /* Load parameters for j particles */
 737             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 738             vdwjidx0A        = 2*vdwtype[jnrA+0];
 739             vdwjidx0B        = 2*vdwtype[jnrB+0];
 740
 741             fjx0             = _mm_setzero_pd();
 742             fjy0             = _mm_setzero_pd();
 743             fjz0             = _mm_setzero_pd();
 744
 745             /**************************
 746              * CALCULATE INTERACTIONS *
 747              **************************/
 748
 749             if (gmx_mm_any_lt(rsq00,rcutoff2))
 750             {
 751
 752             /* Compute parameters for interactions between i and j atoms */
 753             qq00             = _mm_mul_pd(iq0,jq0);
 754             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 755                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 756
 757             /* REACTION-FIELD ELECTROSTATICS */
 758             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 759
 760             /* LENNARD-JONES DISPERSION/REPULSION */
 761
 762             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 763             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 764
 765             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 766
 767             fscal            = _mm_add_pd(felec,fvdw);
 768
 769             fscal            = _mm_and_pd(fscal,cutoff_mask);
 770
 771             /* Calculate temporary vectorial force */
 772             tx               = _mm_mul_pd(fscal,dx00);
 773             ty               = _mm_mul_pd(fscal,dy00);
 774             tz               = _mm_mul_pd(fscal,dz00);
 775
 776             /* Update vectorial force */
 777             fix0             = _mm_add_pd(fix0,tx);
 778             fiy0             = _mm_add_pd(fiy0,ty);
 779             fiz0             = _mm_add_pd(fiz0,tz);
 780
 781             fjx0             = _mm_add_pd(fjx0,tx);
 782             fjy0             = _mm_add_pd(fjy0,ty);
 783             fjz0             = _mm_add_pd(fjz0,tz);
 784
 785             }
 786
 787             /**************************
 788              * CALCULATE INTERACTIONS *
 789              **************************/
 790
 791             if (gmx_mm_any_lt(rsq10,rcutoff2))
 792             {
 793
 794             /* Compute parameters for interactions between i and j atoms */
 795             qq10             = _mm_mul_pd(iq1,jq0);
 796
 797             /* REACTION-FIELD ELECTROSTATICS */
 798             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 799
 800             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 801
 802             fscal            = felec;
 803
 804             fscal            = _mm_and_pd(fscal,cutoff_mask);
 805
 806             /* Calculate temporary vectorial force */
 807             tx               = _mm_mul_pd(fscal,dx10);
 808             ty               = _mm_mul_pd(fscal,dy10);
 809             tz               = _mm_mul_pd(fscal,dz10);
 810
 811             /* Update vectorial force */
 812             fix1             = _mm_add_pd(fix1,tx);
 813             fiy1             = _mm_add_pd(fiy1,ty);
 814             fiz1             = _mm_add_pd(fiz1,tz);
 815
 816             fjx0             = _mm_add_pd(fjx0,tx);
 817             fjy0             = _mm_add_pd(fjy0,ty);
 818             fjz0             = _mm_add_pd(fjz0,tz);
 819
 820             }
 821
 822             /**************************
 823              * CALCULATE INTERACTIONS *
 824              **************************/
 825
 826             if (gmx_mm_any_lt(rsq20,rcutoff2))
 827             {
 828
 829             /* Compute parameters for interactions between i and j atoms */
 830             qq20             = _mm_mul_pd(iq2,jq0);
 831
 832             /* REACTION-FIELD ELECTROSTATICS */
 833             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 834
 835             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 836
 837             fscal            = felec;
 838
 839             fscal            = _mm_and_pd(fscal,cutoff_mask);
 840
 841             /* Calculate temporary vectorial force */
 842             tx               = _mm_mul_pd(fscal,dx20);
 843             ty               = _mm_mul_pd(fscal,dy20);
 844             tz               = _mm_mul_pd(fscal,dz20);
 845
 846             /* Update vectorial force */
 847             fix2             = _mm_add_pd(fix2,tx);
 848             fiy2             = _mm_add_pd(fiy2,ty);
 849             fiz2             = _mm_add_pd(fiz2,tz);
 850
 851             fjx0             = _mm_add_pd(fjx0,tx);
 852             fjy0             = _mm_add_pd(fjy0,ty);
 853             fjz0             = _mm_add_pd(fjz0,tz);
 854
 855             }
 856
 857             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 858
 859             /* Inner loop uses 100 flops */
 860         }
 861
 862         if(jidx<j_index_end)
 863         {
 864
 865             jnrA             = jjnr[jidx];
 866             j_coord_offsetA  = DIM*jnrA;
 867
 868             /* load j atom coordinates */
 869             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 870                                               &jx0,&jy0,&jz0);
 871
 872             /* Calculate displacement vector */
 873             dx00             = _mm_sub_pd(ix0,jx0);
 874             dy00             = _mm_sub_pd(iy0,jy0);
 875             dz00             = _mm_sub_pd(iz0,jz0);
 876             dx10             = _mm_sub_pd(ix1,jx0);
 877             dy10             = _mm_sub_pd(iy1,jy0);
 878             dz10             = _mm_sub_pd(iz1,jz0);
 879             dx20             = _mm_sub_pd(ix2,jx0);
 880             dy20             = _mm_sub_pd(iy2,jy0);
 881             dz20             = _mm_sub_pd(iz2,jz0);
 882
 883             /* Calculate squared distance and things based on it */
 884             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 885             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 886             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 887
 888             rinv00           = sse41_invsqrt_d(rsq00);
 889             rinv10           = sse41_invsqrt_d(rsq10);
 890             rinv20           = sse41_invsqrt_d(rsq20);
 891
 892             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 893             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 894             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 895
 896             /* Load parameters for j particles */
 897             jq0              = _mm_load_sd(charge+jnrA+0);
 898             vdwjidx0A        = 2*vdwtype[jnrA+0];
 899
 900             fjx0             = _mm_setzero_pd();
 901             fjy0             = _mm_setzero_pd();
 902             fjz0             = _mm_setzero_pd();
 903
 904             /**************************
 905              * CALCULATE INTERACTIONS *
 906              **************************/
 907
 908             if (gmx_mm_any_lt(rsq00,rcutoff2))
 909             {
 910
 911             /* Compute parameters for interactions between i and j atoms */
 912             qq00             = _mm_mul_pd(iq0,jq0);
 913             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 914
 915             /* REACTION-FIELD ELECTROSTATICS */
 916             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 917
 918             /* LENNARD-JONES DISPERSION/REPULSION */
 919
 920             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 921             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 922
 923             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 924
 925             fscal            = _mm_add_pd(felec,fvdw);
 926
 927             fscal            = _mm_and_pd(fscal,cutoff_mask);
 928
 929             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 930
 931             /* Calculate temporary vectorial force */
 932             tx               = _mm_mul_pd(fscal,dx00);
 933             ty               = _mm_mul_pd(fscal,dy00);
 934             tz               = _mm_mul_pd(fscal,dz00);
 935
 936             /* Update vectorial force */
 937             fix0             = _mm_add_pd(fix0,tx);
 938             fiy0             = _mm_add_pd(fiy0,ty);
 939             fiz0             = _mm_add_pd(fiz0,tz);
 940
 941             fjx0             = _mm_add_pd(fjx0,tx);
 942             fjy0             = _mm_add_pd(fjy0,ty);
 943             fjz0             = _mm_add_pd(fjz0,tz);
 944
 945             }
 946
 947             /**************************
 948              * CALCULATE INTERACTIONS *
 949              **************************/
 950
 951             if (gmx_mm_any_lt(rsq10,rcutoff2))
 952             {
 953
 954             /* Compute parameters for interactions between i and j atoms */
 955             qq10             = _mm_mul_pd(iq1,jq0);
 956
 957             /* REACTION-FIELD ELECTROSTATICS */
 958             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 959
 960             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 961
 962             fscal            = felec;
 963
 964             fscal            = _mm_and_pd(fscal,cutoff_mask);
 965
 966             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 967
 968             /* Calculate temporary vectorial force */
 969             tx               = _mm_mul_pd(fscal,dx10);
 970             ty               = _mm_mul_pd(fscal,dy10);
 971             tz               = _mm_mul_pd(fscal,dz10);
 972
 973             /* Update vectorial force */
 974             fix1             = _mm_add_pd(fix1,tx);
 975             fiy1             = _mm_add_pd(fiy1,ty);
 976             fiz1             = _mm_add_pd(fiz1,tz);
 977
 978             fjx0             = _mm_add_pd(fjx0,tx);
 979             fjy0             = _mm_add_pd(fjy0,ty);
 980             fjz0             = _mm_add_pd(fjz0,tz);
 981
 982             }
 983
 984             /**************************
 985              * CALCULATE INTERACTIONS *
 986              **************************/
 987
 988             if (gmx_mm_any_lt(rsq20,rcutoff2))
 989             {
 990
 991             /* Compute parameters for interactions between i and j atoms */
 992             qq20             = _mm_mul_pd(iq2,jq0);
 993
 994             /* REACTION-FIELD ELECTROSTATICS */
 995             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 996
 997             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 998
 999             fscal            = felec;
1000
1001             fscal            = _mm_and_pd(fscal,cutoff_mask);
1002
1003             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1004
1005             /* Calculate temporary vectorial force */
1006             tx               = _mm_mul_pd(fscal,dx20);
1007             ty               = _mm_mul_pd(fscal,dy20);
1008             tz               = _mm_mul_pd(fscal,dz20);
1009
1010             /* Update vectorial force */
1011             fix2             = _mm_add_pd(fix2,tx);
1012             fiy2             = _mm_add_pd(fiy2,ty);
1013             fiz2             = _mm_add_pd(fiz2,tz);
1014
1015             fjx0             = _mm_add_pd(fjx0,tx);
1016             fjy0             = _mm_add_pd(fjy0,ty);
1017             fjz0             = _mm_add_pd(fjz0,tz);
1018
1019             }
1020
1021             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1022
1023             /* Inner loop uses 100 flops */
1024         }
1025
1026         /* End of innermost loop */
1027
1028         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1029                                               f+i_coord_offset,fshift+i_shift_offset);
1030
1031         /* Increment number of inner iterations */
1032         inneriter                  += j_index_end - j_index_start;
1033
1034         /* Outer loop uses 18 flops */
1035     }
1036
1037     /* Increment number of outer iterations */
1038     outeriter        += nri;
1039
1040     /* Update outer/inner flops */
1041
1042     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*100);
1043 }