src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_sse2_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_double.h"
  48 #include "kernelutil_x86_sse2_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_VF_sse2_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_VF_sse2_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwjidx0A,vdwjidx0B;
  87     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  89     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  90     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  91     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     int              nvdwtype;
  94     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  95     int              *vdwtype;
  96     real             *vdwparam;
  97     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  98     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  99     __m128d          dummy_mask,cutoff_mask;
 100     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 101     __m128d          one     = _mm_set1_pd(1.0);
 102     __m128d          two     = _mm_set1_pd(2.0);
 103     x                = xx[0];
 104     f                = ff[0];
 105
 106     nri              = nlist->nri;
 107     iinr             = nlist->iinr;
 108     jindex           = nlist->jindex;
 109     jjnr             = nlist->jjnr;
 110     shiftidx         = nlist->shift;
 111     gid              = nlist->gid;
 112     shiftvec         = fr->shift_vec[0];
 113     fshift           = fr->fshift[0];
 114     facel            = _mm_set1_pd(fr->epsfac);
 115     charge           = mdatoms->chargeA;
 116     krf              = _mm_set1_pd(fr->ic->k_rf);
 117     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 118     crf              = _mm_set1_pd(fr->ic->c_rf);
 119     nvdwtype         = fr->ntype;
 120     vdwparam         = fr->nbfp;
 121     vdwtype          = mdatoms->typeA;
 122
 123     /* Setup water-specific parameters */
 124     inr              = nlist->iinr[0];
 125     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 126     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 127     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 128     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 129
 130     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 131     rcutoff_scalar   = fr->rcoulomb;
 132     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 133     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 134
 135     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 136     rvdw             = _mm_set1_pd(fr->rvdw);
 137
 138     /* Avoid stupid compiler warnings */
 139     jnrA = jnrB = 0;
 140     j_coord_offsetA = 0;
 141     j_coord_offsetB = 0;
 142
 143     outeriter        = 0;
 144     inneriter        = 0;
 145
 146     /* Start outer loop over neighborlists */
 147     for(iidx=0; iidx<nri; iidx++)
 148     {
 149         /* Load shift vector for this list */
 150         i_shift_offset   = DIM*shiftidx[iidx];
 151
 152         /* Load limits for loop over neighbors */
 153         j_index_start    = jindex[iidx];
 154         j_index_end      = jindex[iidx+1];
 155
 156         /* Get outer coordinate index */
 157         inr              = iinr[iidx];
 158         i_coord_offset   = DIM*inr;
 159
 160         /* Load i particle coords and add shift vector */
 161         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 162                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 163
 164         fix0             = _mm_setzero_pd();
 165         fiy0             = _mm_setzero_pd();
 166         fiz0             = _mm_setzero_pd();
 167         fix1             = _mm_setzero_pd();
 168         fiy1             = _mm_setzero_pd();
 169         fiz1             = _mm_setzero_pd();
 170         fix2             = _mm_setzero_pd();
 171         fiy2             = _mm_setzero_pd();
 172         fiz2             = _mm_setzero_pd();
 173
 174         /* Reset potential sums */
 175         velecsum         = _mm_setzero_pd();
 176         vvdwsum          = _mm_setzero_pd();
 177
 178         /* Start inner kernel loop */
 179         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 180         {
 181
 182             /* Get j neighbor index, and coordinate index */
 183             jnrA             = jjnr[jidx];
 184             jnrB             = jjnr[jidx+1];
 185             j_coord_offsetA  = DIM*jnrA;
 186             j_coord_offsetB  = DIM*jnrB;
 187
 188             /* load j atom coordinates */
 189             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 190                                               &jx0,&jy0,&jz0);
 191
 192             /* Calculate displacement vector */
 193             dx00             = _mm_sub_pd(ix0,jx0);
 194             dy00             = _mm_sub_pd(iy0,jy0);
 195             dz00             = _mm_sub_pd(iz0,jz0);
 196             dx10             = _mm_sub_pd(ix1,jx0);
 197             dy10             = _mm_sub_pd(iy1,jy0);
 198             dz10             = _mm_sub_pd(iz1,jz0);
 199             dx20             = _mm_sub_pd(ix2,jx0);
 200             dy20             = _mm_sub_pd(iy2,jy0);
 201             dz20             = _mm_sub_pd(iz2,jz0);
 202
 203             /* Calculate squared distance and things based on it */
 204             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 205             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 206             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 207
 208             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 209             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 210             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 211
 212             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 213             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 214             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 215
 216             /* Load parameters for j particles */
 217             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 218             vdwjidx0A        = 2*vdwtype[jnrA+0];
 219             vdwjidx0B        = 2*vdwtype[jnrB+0];
 220
 221             fjx0             = _mm_setzero_pd();
 222             fjy0             = _mm_setzero_pd();
 223             fjz0             = _mm_setzero_pd();
 224
 225             /**************************
 226              * CALCULATE INTERACTIONS *
 227              **************************/
 228
 229             if (gmx_mm_any_lt(rsq00,rcutoff2))
 230             {
 231
 232             /* Compute parameters for interactions between i and j atoms */
 233             qq00             = _mm_mul_pd(iq0,jq0);
 234             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 235                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 236
 237             /* REACTION-FIELD ELECTROSTATICS */
 238             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 239             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 240
 241             /* LENNARD-JONES DISPERSION/REPULSION */
 242
 243             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 244             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 245             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 246             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) ,
 247                                           _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 248             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 249
 250             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 251
 252             /* Update potential sum for this i atom from the interaction with this j atom. */
 253             velec            = _mm_and_pd(velec,cutoff_mask);
 254             velecsum         = _mm_add_pd(velecsum,velec);
 255             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 256             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 257
 258             fscal            = _mm_add_pd(felec,fvdw);
 259
 260             fscal            = _mm_and_pd(fscal,cutoff_mask);
 261
 262             /* Calculate temporary vectorial force */
 263             tx               = _mm_mul_pd(fscal,dx00);
 264             ty               = _mm_mul_pd(fscal,dy00);
 265             tz               = _mm_mul_pd(fscal,dz00);
 266
 267             /* Update vectorial force */
 268             fix0             = _mm_add_pd(fix0,tx);
 269             fiy0             = _mm_add_pd(fiy0,ty);
 270             fiz0             = _mm_add_pd(fiz0,tz);
 271
 272             fjx0             = _mm_add_pd(fjx0,tx);
 273             fjy0             = _mm_add_pd(fjy0,ty);
 274             fjz0             = _mm_add_pd(fjz0,tz);
 275
 276             }
 277
 278             /**************************
 279              * CALCULATE INTERACTIONS *
 280              **************************/
 281
 282             if (gmx_mm_any_lt(rsq10,rcutoff2))
 283             {
 284
 285             /* Compute parameters for interactions between i and j atoms */
 286             qq10             = _mm_mul_pd(iq1,jq0);
 287
 288             /* REACTION-FIELD ELECTROSTATICS */
 289             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 290             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 291
 292             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 293
 294             /* Update potential sum for this i atom from the interaction with this j atom. */
 295             velec            = _mm_and_pd(velec,cutoff_mask);
 296             velecsum         = _mm_add_pd(velecsum,velec);
 297
 298             fscal            = felec;
 299
 300             fscal            = _mm_and_pd(fscal,cutoff_mask);
 301
 302             /* Calculate temporary vectorial force */
 303             tx               = _mm_mul_pd(fscal,dx10);
 304             ty               = _mm_mul_pd(fscal,dy10);
 305             tz               = _mm_mul_pd(fscal,dz10);
 306
 307             /* Update vectorial force */
 308             fix1             = _mm_add_pd(fix1,tx);
 309             fiy1             = _mm_add_pd(fiy1,ty);
 310             fiz1             = _mm_add_pd(fiz1,tz);
 311
 312             fjx0             = _mm_add_pd(fjx0,tx);
 313             fjy0             = _mm_add_pd(fjy0,ty);
 314             fjz0             = _mm_add_pd(fjz0,tz);
 315
 316             }
 317
 318             /**************************
 319              * CALCULATE INTERACTIONS *
 320              **************************/
 321
 322             if (gmx_mm_any_lt(rsq20,rcutoff2))
 323             {
 324
 325             /* Compute parameters for interactions between i and j atoms */
 326             qq20             = _mm_mul_pd(iq2,jq0);
 327
 328             /* REACTION-FIELD ELECTROSTATICS */
 329             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 330             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 331
 332             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 333
 334             /* Update potential sum for this i atom from the interaction with this j atom. */
 335             velec            = _mm_and_pd(velec,cutoff_mask);
 336             velecsum         = _mm_add_pd(velecsum,velec);
 337
 338             fscal            = felec;
 339
 340             fscal            = _mm_and_pd(fscal,cutoff_mask);
 341
 342             /* Calculate temporary vectorial force */
 343             tx               = _mm_mul_pd(fscal,dx20);
 344             ty               = _mm_mul_pd(fscal,dy20);
 345             tz               = _mm_mul_pd(fscal,dz20);
 346
 347             /* Update vectorial force */
 348             fix2             = _mm_add_pd(fix2,tx);
 349             fiy2             = _mm_add_pd(fiy2,ty);
 350             fiz2             = _mm_add_pd(fiz2,tz);
 351
 352             fjx0             = _mm_add_pd(fjx0,tx);
 353             fjy0             = _mm_add_pd(fjy0,ty);
 354             fjz0             = _mm_add_pd(fjz0,tz);
 355
 356             }
 357
 358             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 359
 360             /* Inner loop uses 129 flops */
 361         }
 362
 363         if(jidx<j_index_end)
 364         {
 365
 366             jnrA             = jjnr[jidx];
 367             j_coord_offsetA  = DIM*jnrA;
 368
 369             /* load j atom coordinates */
 370             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 371                                               &jx0,&jy0,&jz0);
 372
 373             /* Calculate displacement vector */
 374             dx00             = _mm_sub_pd(ix0,jx0);
 375             dy00             = _mm_sub_pd(iy0,jy0);
 376             dz00             = _mm_sub_pd(iz0,jz0);
 377             dx10             = _mm_sub_pd(ix1,jx0);
 378             dy10             = _mm_sub_pd(iy1,jy0);
 379             dz10             = _mm_sub_pd(iz1,jz0);
 380             dx20             = _mm_sub_pd(ix2,jx0);
 381             dy20             = _mm_sub_pd(iy2,jy0);
 382             dz20             = _mm_sub_pd(iz2,jz0);
 383
 384             /* Calculate squared distance and things based on it */
 385             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 386             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 387             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 388
 389             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 390             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 391             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 392
 393             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 394             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 395             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 396
 397             /* Load parameters for j particles */
 398             jq0              = _mm_load_sd(charge+jnrA+0);
 399             vdwjidx0A        = 2*vdwtype[jnrA+0];
 400
 401             fjx0             = _mm_setzero_pd();
 402             fjy0             = _mm_setzero_pd();
 403             fjz0             = _mm_setzero_pd();
 404
 405             /**************************
 406              * CALCULATE INTERACTIONS *
 407              **************************/
 408
 409             if (gmx_mm_any_lt(rsq00,rcutoff2))
 410             {
 411
 412             /* Compute parameters for interactions between i and j atoms */
 413             qq00             = _mm_mul_pd(iq0,jq0);
 414             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 415
 416             /* REACTION-FIELD ELECTROSTATICS */
 417             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 418             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 419
 420             /* LENNARD-JONES DISPERSION/REPULSION */
 421
 422             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 423             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 424             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 425             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) ,
 426                                           _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 427             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 428
 429             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 430
 431             /* Update potential sum for this i atom from the interaction with this j atom. */
 432             velec            = _mm_and_pd(velec,cutoff_mask);
 433             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 434             velecsum         = _mm_add_pd(velecsum,velec);
 435             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 436             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 437             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 438
 439             fscal            = _mm_add_pd(felec,fvdw);
 440
 441             fscal            = _mm_and_pd(fscal,cutoff_mask);
 442
 443             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 444
 445             /* Calculate temporary vectorial force */
 446             tx               = _mm_mul_pd(fscal,dx00);
 447             ty               = _mm_mul_pd(fscal,dy00);
 448             tz               = _mm_mul_pd(fscal,dz00);
 449
 450             /* Update vectorial force */
 451             fix0             = _mm_add_pd(fix0,tx);
 452             fiy0             = _mm_add_pd(fiy0,ty);
 453             fiz0             = _mm_add_pd(fiz0,tz);
 454
 455             fjx0             = _mm_add_pd(fjx0,tx);
 456             fjy0             = _mm_add_pd(fjy0,ty);
 457             fjz0             = _mm_add_pd(fjz0,tz);
 458
 459             }
 460
 461             /**************************
 462              * CALCULATE INTERACTIONS *
 463              **************************/
 464
 465             if (gmx_mm_any_lt(rsq10,rcutoff2))
 466             {
 467
 468             /* Compute parameters for interactions between i and j atoms */
 469             qq10             = _mm_mul_pd(iq1,jq0);
 470
 471             /* REACTION-FIELD ELECTROSTATICS */
 472             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 473             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 474
 475             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 476
 477             /* Update potential sum for this i atom from the interaction with this j atom. */
 478             velec            = _mm_and_pd(velec,cutoff_mask);
 479             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 480             velecsum         = _mm_add_pd(velecsum,velec);
 481
 482             fscal            = felec;
 483
 484             fscal            = _mm_and_pd(fscal,cutoff_mask);
 485
 486             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 487
 488             /* Calculate temporary vectorial force */
 489             tx               = _mm_mul_pd(fscal,dx10);
 490             ty               = _mm_mul_pd(fscal,dy10);
 491             tz               = _mm_mul_pd(fscal,dz10);
 492
 493             /* Update vectorial force */
 494             fix1             = _mm_add_pd(fix1,tx);
 495             fiy1             = _mm_add_pd(fiy1,ty);
 496             fiz1             = _mm_add_pd(fiz1,tz);
 497
 498             fjx0             = _mm_add_pd(fjx0,tx);
 499             fjy0             = _mm_add_pd(fjy0,ty);
 500             fjz0             = _mm_add_pd(fjz0,tz);
 501
 502             }
 503
 504             /**************************
 505              * CALCULATE INTERACTIONS *
 506              **************************/
 507
 508             if (gmx_mm_any_lt(rsq20,rcutoff2))
 509             {
 510
 511             /* Compute parameters for interactions between i and j atoms */
 512             qq20             = _mm_mul_pd(iq2,jq0);
 513
 514             /* REACTION-FIELD ELECTROSTATICS */
 515             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 516             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 517
 518             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 519
 520             /* Update potential sum for this i atom from the interaction with this j atom. */
 521             velec            = _mm_and_pd(velec,cutoff_mask);
 522             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 523             velecsum         = _mm_add_pd(velecsum,velec);
 524
 525             fscal            = felec;
 526
 527             fscal            = _mm_and_pd(fscal,cutoff_mask);
 528
 529             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 530
 531             /* Calculate temporary vectorial force */
 532             tx               = _mm_mul_pd(fscal,dx20);
 533             ty               = _mm_mul_pd(fscal,dy20);
 534             tz               = _mm_mul_pd(fscal,dz20);
 535
 536             /* Update vectorial force */
 537             fix2             = _mm_add_pd(fix2,tx);
 538             fiy2             = _mm_add_pd(fiy2,ty);
 539             fiz2             = _mm_add_pd(fiz2,tz);
 540
 541             fjx0             = _mm_add_pd(fjx0,tx);
 542             fjy0             = _mm_add_pd(fjy0,ty);
 543             fjz0             = _mm_add_pd(fjz0,tz);
 544
 545             }
 546
 547             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 548
 549             /* Inner loop uses 129 flops */
 550         }
 551
 552         /* End of innermost loop */
 553
 554         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 555                                               f+i_coord_offset,fshift+i_shift_offset);
 556
 557         ggid                        = gid[iidx];
 558         /* Update potential energies */
 559         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 560         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 561
 562         /* Increment number of inner iterations */
 563         inneriter                  += j_index_end - j_index_start;
 564
 565         /* Outer loop uses 20 flops */
 566     }
 567
 568     /* Increment number of outer iterations */
 569     outeriter        += nri;
 570
 571     /* Update outer/inner flops */
 572
 573     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*129);
 574 }
 575 /*
 576  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_sse2_double
 577  * Electrostatics interaction: ReactionField
 578  * VdW interaction:            LennardJones
 579  * Geometry:                   Water3-Particle
 580  * Calculate force/pot:        Force
 581  */
 582 void
 583 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_sse2_double
 584                     (t_nblist                    * gmx_restrict       nlist,
 585                      rvec                        * gmx_restrict          xx,
 586                      rvec                        * gmx_restrict          ff,
 587                      t_forcerec                  * gmx_restrict          fr,
 588                      t_mdatoms                   * gmx_restrict     mdatoms,
 589                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 590                      t_nrnb                      * gmx_restrict        nrnb)
 591 {
 592     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 593      * just 0 for non-waters.
 594      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 595      * jnr indices corresponding to data put in the four positions in the SIMD register.
 596      */
 597     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 598     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 599     int              jnrA,jnrB;
 600     int              j_coord_offsetA,j_coord_offsetB;
 601     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 602     real             rcutoff_scalar;
 603     real             *shiftvec,*fshift,*x,*f;
 604     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 605     int              vdwioffset0;
 606     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 607     int              vdwioffset1;
 608     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 609     int              vdwioffset2;
 610     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 611     int              vdwjidx0A,vdwjidx0B;
 612     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 613     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 614     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 615     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 616     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 617     real             *charge;
 618     int              nvdwtype;
 619     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 620     int              *vdwtype;
 621     real             *vdwparam;
 622     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 623     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 624     __m128d          dummy_mask,cutoff_mask;
 625     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 626     __m128d          one     = _mm_set1_pd(1.0);
 627     __m128d          two     = _mm_set1_pd(2.0);
 628     x                = xx[0];
 629     f                = ff[0];
 630
 631     nri              = nlist->nri;
 632     iinr             = nlist->iinr;
 633     jindex           = nlist->jindex;
 634     jjnr             = nlist->jjnr;
 635     shiftidx         = nlist->shift;
 636     gid              = nlist->gid;
 637     shiftvec         = fr->shift_vec[0];
 638     fshift           = fr->fshift[0];
 639     facel            = _mm_set1_pd(fr->epsfac);
 640     charge           = mdatoms->chargeA;
 641     krf              = _mm_set1_pd(fr->ic->k_rf);
 642     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 643     crf              = _mm_set1_pd(fr->ic->c_rf);
 644     nvdwtype         = fr->ntype;
 645     vdwparam         = fr->nbfp;
 646     vdwtype          = mdatoms->typeA;
 647
 648     /* Setup water-specific parameters */
 649     inr              = nlist->iinr[0];
 650     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 651     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 652     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 653     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 654
 655     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 656     rcutoff_scalar   = fr->rcoulomb;
 657     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 658     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 659
 660     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 661     rvdw             = _mm_set1_pd(fr->rvdw);
 662
 663     /* Avoid stupid compiler warnings */
 664     jnrA = jnrB = 0;
 665     j_coord_offsetA = 0;
 666     j_coord_offsetB = 0;
 667
 668     outeriter        = 0;
 669     inneriter        = 0;
 670
 671     /* Start outer loop over neighborlists */
 672     for(iidx=0; iidx<nri; iidx++)
 673     {
 674         /* Load shift vector for this list */
 675         i_shift_offset   = DIM*shiftidx[iidx];
 676
 677         /* Load limits for loop over neighbors */
 678         j_index_start    = jindex[iidx];
 679         j_index_end      = jindex[iidx+1];
 680
 681         /* Get outer coordinate index */
 682         inr              = iinr[iidx];
 683         i_coord_offset   = DIM*inr;
 684
 685         /* Load i particle coords and add shift vector */
 686         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 687                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 688
 689         fix0             = _mm_setzero_pd();
 690         fiy0             = _mm_setzero_pd();
 691         fiz0             = _mm_setzero_pd();
 692         fix1             = _mm_setzero_pd();
 693         fiy1             = _mm_setzero_pd();
 694         fiz1             = _mm_setzero_pd();
 695         fix2             = _mm_setzero_pd();
 696         fiy2             = _mm_setzero_pd();
 697         fiz2             = _mm_setzero_pd();
 698
 699         /* Start inner kernel loop */
 700         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 701         {
 702
 703             /* Get j neighbor index, and coordinate index */
 704             jnrA             = jjnr[jidx];
 705             jnrB             = jjnr[jidx+1];
 706             j_coord_offsetA  = DIM*jnrA;
 707             j_coord_offsetB  = DIM*jnrB;
 708
 709             /* load j atom coordinates */
 710             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 711                                               &jx0,&jy0,&jz0);
 712
 713             /* Calculate displacement vector */
 714             dx00             = _mm_sub_pd(ix0,jx0);
 715             dy00             = _mm_sub_pd(iy0,jy0);
 716             dz00             = _mm_sub_pd(iz0,jz0);
 717             dx10             = _mm_sub_pd(ix1,jx0);
 718             dy10             = _mm_sub_pd(iy1,jy0);
 719             dz10             = _mm_sub_pd(iz1,jz0);
 720             dx20             = _mm_sub_pd(ix2,jx0);
 721             dy20             = _mm_sub_pd(iy2,jy0);
 722             dz20             = _mm_sub_pd(iz2,jz0);
 723
 724             /* Calculate squared distance and things based on it */
 725             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 726             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 727             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 728
 729             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 730             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 731             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 732
 733             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 734             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 735             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 736
 737             /* Load parameters for j particles */
 738             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 739             vdwjidx0A        = 2*vdwtype[jnrA+0];
 740             vdwjidx0B        = 2*vdwtype[jnrB+0];
 741
 742             fjx0             = _mm_setzero_pd();
 743             fjy0             = _mm_setzero_pd();
 744             fjz0             = _mm_setzero_pd();
 745
 746             /**************************
 747              * CALCULATE INTERACTIONS *
 748              **************************/
 749
 750             if (gmx_mm_any_lt(rsq00,rcutoff2))
 751             {
 752
 753             /* Compute parameters for interactions between i and j atoms */
 754             qq00             = _mm_mul_pd(iq0,jq0);
 755             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 756                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 757
 758             /* REACTION-FIELD ELECTROSTATICS */
 759             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 760
 761             /* LENNARD-JONES DISPERSION/REPULSION */
 762
 763             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 764             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 765
 766             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 767
 768             fscal            = _mm_add_pd(felec,fvdw);
 769
 770             fscal            = _mm_and_pd(fscal,cutoff_mask);
 771
 772             /* Calculate temporary vectorial force */
 773             tx               = _mm_mul_pd(fscal,dx00);
 774             ty               = _mm_mul_pd(fscal,dy00);
 775             tz               = _mm_mul_pd(fscal,dz00);
 776
 777             /* Update vectorial force */
 778             fix0             = _mm_add_pd(fix0,tx);
 779             fiy0             = _mm_add_pd(fiy0,ty);
 780             fiz0             = _mm_add_pd(fiz0,tz);
 781
 782             fjx0             = _mm_add_pd(fjx0,tx);
 783             fjy0             = _mm_add_pd(fjy0,ty);
 784             fjz0             = _mm_add_pd(fjz0,tz);
 785
 786             }
 787
 788             /**************************
 789              * CALCULATE INTERACTIONS *
 790              **************************/
 791
 792             if (gmx_mm_any_lt(rsq10,rcutoff2))
 793             {
 794
 795             /* Compute parameters for interactions between i and j atoms */
 796             qq10             = _mm_mul_pd(iq1,jq0);
 797
 798             /* REACTION-FIELD ELECTROSTATICS */
 799             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 800
 801             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 802
 803             fscal            = felec;
 804
 805             fscal            = _mm_and_pd(fscal,cutoff_mask);
 806
 807             /* Calculate temporary vectorial force */
 808             tx               = _mm_mul_pd(fscal,dx10);
 809             ty               = _mm_mul_pd(fscal,dy10);
 810             tz               = _mm_mul_pd(fscal,dz10);
 811
 812             /* Update vectorial force */
 813             fix1             = _mm_add_pd(fix1,tx);
 814             fiy1             = _mm_add_pd(fiy1,ty);
 815             fiz1             = _mm_add_pd(fiz1,tz);
 816
 817             fjx0             = _mm_add_pd(fjx0,tx);
 818             fjy0             = _mm_add_pd(fjy0,ty);
 819             fjz0             = _mm_add_pd(fjz0,tz);
 820
 821             }
 822
 823             /**************************
 824              * CALCULATE INTERACTIONS *
 825              **************************/
 826
 827             if (gmx_mm_any_lt(rsq20,rcutoff2))
 828             {
 829
 830             /* Compute parameters for interactions between i and j atoms */
 831             qq20             = _mm_mul_pd(iq2,jq0);
 832
 833             /* REACTION-FIELD ELECTROSTATICS */
 834             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 835
 836             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 837
 838             fscal            = felec;
 839
 840             fscal            = _mm_and_pd(fscal,cutoff_mask);
 841
 842             /* Calculate temporary vectorial force */
 843             tx               = _mm_mul_pd(fscal,dx20);
 844             ty               = _mm_mul_pd(fscal,dy20);
 845             tz               = _mm_mul_pd(fscal,dz20);
 846
 847             /* Update vectorial force */
 848             fix2             = _mm_add_pd(fix2,tx);
 849             fiy2             = _mm_add_pd(fiy2,ty);
 850             fiz2             = _mm_add_pd(fiz2,tz);
 851
 852             fjx0             = _mm_add_pd(fjx0,tx);
 853             fjy0             = _mm_add_pd(fjy0,ty);
 854             fjz0             = _mm_add_pd(fjz0,tz);
 855
 856             }
 857
 858             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 859
 860             /* Inner loop uses 100 flops */
 861         }
 862
 863         if(jidx<j_index_end)
 864         {
 865
 866             jnrA             = jjnr[jidx];
 867             j_coord_offsetA  = DIM*jnrA;
 868
 869             /* load j atom coordinates */
 870             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 871                                               &jx0,&jy0,&jz0);
 872
 873             /* Calculate displacement vector */
 874             dx00             = _mm_sub_pd(ix0,jx0);
 875             dy00             = _mm_sub_pd(iy0,jy0);
 876             dz00             = _mm_sub_pd(iz0,jz0);
 877             dx10             = _mm_sub_pd(ix1,jx0);
 878             dy10             = _mm_sub_pd(iy1,jy0);
 879             dz10             = _mm_sub_pd(iz1,jz0);
 880             dx20             = _mm_sub_pd(ix2,jx0);
 881             dy20             = _mm_sub_pd(iy2,jy0);
 882             dz20             = _mm_sub_pd(iz2,jz0);
 883
 884             /* Calculate squared distance and things based on it */
 885             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 886             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 887             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 888
 889             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 890             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 891             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 892
 893             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 894             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 895             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 896
 897             /* Load parameters for j particles */
 898             jq0              = _mm_load_sd(charge+jnrA+0);
 899             vdwjidx0A        = 2*vdwtype[jnrA+0];
 900
 901             fjx0             = _mm_setzero_pd();
 902             fjy0             = _mm_setzero_pd();
 903             fjz0             = _mm_setzero_pd();
 904
 905             /**************************
 906              * CALCULATE INTERACTIONS *
 907              **************************/
 908
 909             if (gmx_mm_any_lt(rsq00,rcutoff2))
 910             {
 911
 912             /* Compute parameters for interactions between i and j atoms */
 913             qq00             = _mm_mul_pd(iq0,jq0);
 914             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 915
 916             /* REACTION-FIELD ELECTROSTATICS */
 917             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 918
 919             /* LENNARD-JONES DISPERSION/REPULSION */
 920
 921             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 922             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 923
 924             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 925
 926             fscal            = _mm_add_pd(felec,fvdw);
 927
 928             fscal            = _mm_and_pd(fscal,cutoff_mask);
 929
 930             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 931
 932             /* Calculate temporary vectorial force */
 933             tx               = _mm_mul_pd(fscal,dx00);
 934             ty               = _mm_mul_pd(fscal,dy00);
 935             tz               = _mm_mul_pd(fscal,dz00);
 936
 937             /* Update vectorial force */
 938             fix0             = _mm_add_pd(fix0,tx);
 939             fiy0             = _mm_add_pd(fiy0,ty);
 940             fiz0             = _mm_add_pd(fiz0,tz);
 941
 942             fjx0             = _mm_add_pd(fjx0,tx);
 943             fjy0             = _mm_add_pd(fjy0,ty);
 944             fjz0             = _mm_add_pd(fjz0,tz);
 945
 946             }
 947
 948             /**************************
 949              * CALCULATE INTERACTIONS *
 950              **************************/
 951
 952             if (gmx_mm_any_lt(rsq10,rcutoff2))
 953             {
 954
 955             /* Compute parameters for interactions between i and j atoms */
 956             qq10             = _mm_mul_pd(iq1,jq0);
 957
 958             /* REACTION-FIELD ELECTROSTATICS */
 959             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 960
 961             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 962
 963             fscal            = felec;
 964
 965             fscal            = _mm_and_pd(fscal,cutoff_mask);
 966
 967             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 968
 969             /* Calculate temporary vectorial force */
 970             tx               = _mm_mul_pd(fscal,dx10);
 971             ty               = _mm_mul_pd(fscal,dy10);
 972             tz               = _mm_mul_pd(fscal,dz10);
 973
 974             /* Update vectorial force */
 975             fix1             = _mm_add_pd(fix1,tx);
 976             fiy1             = _mm_add_pd(fiy1,ty);
 977             fiz1             = _mm_add_pd(fiz1,tz);
 978
 979             fjx0             = _mm_add_pd(fjx0,tx);
 980             fjy0             = _mm_add_pd(fjy0,ty);
 981             fjz0             = _mm_add_pd(fjz0,tz);
 982
 983             }
 984
 985             /**************************
 986              * CALCULATE INTERACTIONS *
 987              **************************/
 988
 989             if (gmx_mm_any_lt(rsq20,rcutoff2))
 990             {
 991
 992             /* Compute parameters for interactions between i and j atoms */
 993             qq20             = _mm_mul_pd(iq2,jq0);
 994
 995             /* REACTION-FIELD ELECTROSTATICS */
 996             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 997
 998             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 999
1000             fscal            = felec;
1001
1002             fscal            = _mm_and_pd(fscal,cutoff_mask);
1003
1004             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1005
1006             /* Calculate temporary vectorial force */
1007             tx               = _mm_mul_pd(fscal,dx20);
1008             ty               = _mm_mul_pd(fscal,dy20);
1009             tz               = _mm_mul_pd(fscal,dz20);
1010
1011             /* Update vectorial force */
1012             fix2             = _mm_add_pd(fix2,tx);
1013             fiy2             = _mm_add_pd(fiy2,ty);
1014             fiz2             = _mm_add_pd(fiz2,tz);
1015
1016             fjx0             = _mm_add_pd(fjx0,tx);
1017             fjy0             = _mm_add_pd(fjy0,ty);
1018             fjz0             = _mm_add_pd(fjz0,tz);
1019
1020             }
1021
1022             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1023
1024             /* Inner loop uses 100 flops */
1025         }
1026
1027         /* End of innermost loop */
1028
1029         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1030                                               f+i_coord_offset,fshift+i_shift_offset);
1031
1032         /* Increment number of inner iterations */
1033         inneriter                  += j_index_end - j_index_start;
1034
1035         /* Outer loop uses 18 flops */
1036     }
1037
1038     /* Increment number of outer iterations */
1039     outeriter        += nri;
1040
1041     /* Update outer/inner flops */
1042
1043     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*100);
1044 }