src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_sse4_1_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 sse4_1_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_double.h"
  48 #include "kernelutil_x86_sse4_1_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_VF_sse4_1_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_VF_sse4_1_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              vdwioffset3;
  87     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  94     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 101     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 102     __m128d          dummy_mask,cutoff_mask;
 103     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 104     __m128d          one     = _mm_set1_pd(1.0);
 105     __m128d          two     = _mm_set1_pd(2.0);
 106     x                = xx[0];
 107     f                = ff[0];
 108
 109     nri              = nlist->nri;
 110     iinr             = nlist->iinr;
 111     jindex           = nlist->jindex;
 112     jjnr             = nlist->jjnr;
 113     shiftidx         = nlist->shift;
 114     gid              = nlist->gid;
 115     shiftvec         = fr->shift_vec[0];
 116     fshift           = fr->fshift[0];
 117     facel            = _mm_set1_pd(fr->epsfac);
 118     charge           = mdatoms->chargeA;
 119     krf              = _mm_set1_pd(fr->ic->k_rf);
 120     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 121     crf              = _mm_set1_pd(fr->ic->c_rf);
 122     nvdwtype         = fr->ntype;
 123     vdwparam         = fr->nbfp;
 124     vdwtype          = mdatoms->typeA;
 125
 126     /* Setup water-specific parameters */
 127     inr              = nlist->iinr[0];
 128     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 129     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 130     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 131     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 132
 133     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 134     rcutoff_scalar   = fr->rcoulomb;
 135     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 136     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 137
 138     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 139     rvdw             = _mm_set1_pd(fr->rvdw);
 140
 141     /* Avoid stupid compiler warnings */
 142     jnrA = jnrB = 0;
 143     j_coord_offsetA = 0;
 144     j_coord_offsetB = 0;
 145
 146     outeriter        = 0;
 147     inneriter        = 0;
 148
 149     /* Start outer loop over neighborlists */
 150     for(iidx=0; iidx<nri; iidx++)
 151     {
 152         /* Load shift vector for this list */
 153         i_shift_offset   = DIM*shiftidx[iidx];
 154
 155         /* Load limits for loop over neighbors */
 156         j_index_start    = jindex[iidx];
 157         j_index_end      = jindex[iidx+1];
 158
 159         /* Get outer coordinate index */
 160         inr              = iinr[iidx];
 161         i_coord_offset   = DIM*inr;
 162
 163         /* Load i particle coords and add shift vector */
 164         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 165                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 166
 167         fix0             = _mm_setzero_pd();
 168         fiy0             = _mm_setzero_pd();
 169         fiz0             = _mm_setzero_pd();
 170         fix1             = _mm_setzero_pd();
 171         fiy1             = _mm_setzero_pd();
 172         fiz1             = _mm_setzero_pd();
 173         fix2             = _mm_setzero_pd();
 174         fiy2             = _mm_setzero_pd();
 175         fiz2             = _mm_setzero_pd();
 176         fix3             = _mm_setzero_pd();
 177         fiy3             = _mm_setzero_pd();
 178         fiz3             = _mm_setzero_pd();
 179
 180         /* Reset potential sums */
 181         velecsum         = _mm_setzero_pd();
 182         vvdwsum          = _mm_setzero_pd();
 183
 184         /* Start inner kernel loop */
 185         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 186         {
 187
 188             /* Get j neighbor index, and coordinate index */
 189             jnrA             = jjnr[jidx];
 190             jnrB             = jjnr[jidx+1];
 191             j_coord_offsetA  = DIM*jnrA;
 192             j_coord_offsetB  = DIM*jnrB;
 193
 194             /* load j atom coordinates */
 195             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 196                                               &jx0,&jy0,&jz0);
 197
 198             /* Calculate displacement vector */
 199             dx00             = _mm_sub_pd(ix0,jx0);
 200             dy00             = _mm_sub_pd(iy0,jy0);
 201             dz00             = _mm_sub_pd(iz0,jz0);
 202             dx10             = _mm_sub_pd(ix1,jx0);
 203             dy10             = _mm_sub_pd(iy1,jy0);
 204             dz10             = _mm_sub_pd(iz1,jz0);
 205             dx20             = _mm_sub_pd(ix2,jx0);
 206             dy20             = _mm_sub_pd(iy2,jy0);
 207             dz20             = _mm_sub_pd(iz2,jz0);
 208             dx30             = _mm_sub_pd(ix3,jx0);
 209             dy30             = _mm_sub_pd(iy3,jy0);
 210             dz30             = _mm_sub_pd(iz3,jz0);
 211
 212             /* Calculate squared distance and things based on it */
 213             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 214             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 215             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 216             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 217
 218             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 219             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 220             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 221
 222             rinvsq00         = gmx_mm_inv_pd(rsq00);
 223             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 224             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 225             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 226
 227             /* Load parameters for j particles */
 228             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 229             vdwjidx0A        = 2*vdwtype[jnrA+0];
 230             vdwjidx0B        = 2*vdwtype[jnrB+0];
 231
 232             fjx0             = _mm_setzero_pd();
 233             fjy0             = _mm_setzero_pd();
 234             fjz0             = _mm_setzero_pd();
 235
 236             /**************************
 237              * CALCULATE INTERACTIONS *
 238              **************************/
 239
 240             if (gmx_mm_any_lt(rsq00,rcutoff2))
 241             {
 242
 243             /* Compute parameters for interactions between i and j atoms */
 244             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 245                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 246
 247             /* LENNARD-JONES DISPERSION/REPULSION */
 248
 249             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 250             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 251             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 252             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) ,
 253                                           _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 254             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 255
 256             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 257
 258             /* Update potential sum for this i atom from the interaction with this j atom. */
 259             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 260             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 261
 262             fscal            = fvdw;
 263
 264             fscal            = _mm_and_pd(fscal,cutoff_mask);
 265
 266             /* Calculate temporary vectorial force */
 267             tx               = _mm_mul_pd(fscal,dx00);
 268             ty               = _mm_mul_pd(fscal,dy00);
 269             tz               = _mm_mul_pd(fscal,dz00);
 270
 271             /* Update vectorial force */
 272             fix0             = _mm_add_pd(fix0,tx);
 273             fiy0             = _mm_add_pd(fiy0,ty);
 274             fiz0             = _mm_add_pd(fiz0,tz);
 275
 276             fjx0             = _mm_add_pd(fjx0,tx);
 277             fjy0             = _mm_add_pd(fjy0,ty);
 278             fjz0             = _mm_add_pd(fjz0,tz);
 279
 280             }
 281
 282             /**************************
 283              * CALCULATE INTERACTIONS *
 284              **************************/
 285
 286             if (gmx_mm_any_lt(rsq10,rcutoff2))
 287             {
 288
 289             /* Compute parameters for interactions between i and j atoms */
 290             qq10             = _mm_mul_pd(iq1,jq0);
 291
 292             /* REACTION-FIELD ELECTROSTATICS */
 293             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 294             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 295
 296             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 297
 298             /* Update potential sum for this i atom from the interaction with this j atom. */
 299             velec            = _mm_and_pd(velec,cutoff_mask);
 300             velecsum         = _mm_add_pd(velecsum,velec);
 301
 302             fscal            = felec;
 303
 304             fscal            = _mm_and_pd(fscal,cutoff_mask);
 305
 306             /* Calculate temporary vectorial force */
 307             tx               = _mm_mul_pd(fscal,dx10);
 308             ty               = _mm_mul_pd(fscal,dy10);
 309             tz               = _mm_mul_pd(fscal,dz10);
 310
 311             /* Update vectorial force */
 312             fix1             = _mm_add_pd(fix1,tx);
 313             fiy1             = _mm_add_pd(fiy1,ty);
 314             fiz1             = _mm_add_pd(fiz1,tz);
 315
 316             fjx0             = _mm_add_pd(fjx0,tx);
 317             fjy0             = _mm_add_pd(fjy0,ty);
 318             fjz0             = _mm_add_pd(fjz0,tz);
 319
 320             }
 321
 322             /**************************
 323              * CALCULATE INTERACTIONS *
 324              **************************/
 325
 326             if (gmx_mm_any_lt(rsq20,rcutoff2))
 327             {
 328
 329             /* Compute parameters for interactions between i and j atoms */
 330             qq20             = _mm_mul_pd(iq2,jq0);
 331
 332             /* REACTION-FIELD ELECTROSTATICS */
 333             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 334             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 335
 336             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 337
 338             /* Update potential sum for this i atom from the interaction with this j atom. */
 339             velec            = _mm_and_pd(velec,cutoff_mask);
 340             velecsum         = _mm_add_pd(velecsum,velec);
 341
 342             fscal            = felec;
 343
 344             fscal            = _mm_and_pd(fscal,cutoff_mask);
 345
 346             /* Calculate temporary vectorial force */
 347             tx               = _mm_mul_pd(fscal,dx20);
 348             ty               = _mm_mul_pd(fscal,dy20);
 349             tz               = _mm_mul_pd(fscal,dz20);
 350
 351             /* Update vectorial force */
 352             fix2             = _mm_add_pd(fix2,tx);
 353             fiy2             = _mm_add_pd(fiy2,ty);
 354             fiz2             = _mm_add_pd(fiz2,tz);
 355
 356             fjx0             = _mm_add_pd(fjx0,tx);
 357             fjy0             = _mm_add_pd(fjy0,ty);
 358             fjz0             = _mm_add_pd(fjz0,tz);
 359
 360             }
 361
 362             /**************************
 363              * CALCULATE INTERACTIONS *
 364              **************************/
 365
 366             if (gmx_mm_any_lt(rsq30,rcutoff2))
 367             {
 368
 369             /* Compute parameters for interactions between i and j atoms */
 370             qq30             = _mm_mul_pd(iq3,jq0);
 371
 372             /* REACTION-FIELD ELECTROSTATICS */
 373             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_add_pd(rinv30,_mm_mul_pd(krf,rsq30)),crf));
 374             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
 375
 376             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 377
 378             /* Update potential sum for this i atom from the interaction with this j atom. */
 379             velec            = _mm_and_pd(velec,cutoff_mask);
 380             velecsum         = _mm_add_pd(velecsum,velec);
 381
 382             fscal            = felec;
 383
 384             fscal            = _mm_and_pd(fscal,cutoff_mask);
 385
 386             /* Calculate temporary vectorial force */
 387             tx               = _mm_mul_pd(fscal,dx30);
 388             ty               = _mm_mul_pd(fscal,dy30);
 389             tz               = _mm_mul_pd(fscal,dz30);
 390
 391             /* Update vectorial force */
 392             fix3             = _mm_add_pd(fix3,tx);
 393             fiy3             = _mm_add_pd(fiy3,ty);
 394             fiz3             = _mm_add_pd(fiz3,tz);
 395
 396             fjx0             = _mm_add_pd(fjx0,tx);
 397             fjy0             = _mm_add_pd(fjy0,ty);
 398             fjz0             = _mm_add_pd(fjz0,tz);
 399
 400             }
 401
 402             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 403
 404             /* Inner loop uses 152 flops */
 405         }
 406
 407         if(jidx<j_index_end)
 408         {
 409
 410             jnrA             = jjnr[jidx];
 411             j_coord_offsetA  = DIM*jnrA;
 412
 413             /* load j atom coordinates */
 414             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 415                                               &jx0,&jy0,&jz0);
 416
 417             /* Calculate displacement vector */
 418             dx00             = _mm_sub_pd(ix0,jx0);
 419             dy00             = _mm_sub_pd(iy0,jy0);
 420             dz00             = _mm_sub_pd(iz0,jz0);
 421             dx10             = _mm_sub_pd(ix1,jx0);
 422             dy10             = _mm_sub_pd(iy1,jy0);
 423             dz10             = _mm_sub_pd(iz1,jz0);
 424             dx20             = _mm_sub_pd(ix2,jx0);
 425             dy20             = _mm_sub_pd(iy2,jy0);
 426             dz20             = _mm_sub_pd(iz2,jz0);
 427             dx30             = _mm_sub_pd(ix3,jx0);
 428             dy30             = _mm_sub_pd(iy3,jy0);
 429             dz30             = _mm_sub_pd(iz3,jz0);
 430
 431             /* Calculate squared distance and things based on it */
 432             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 433             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 434             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 435             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 436
 437             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 438             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 439             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 440
 441             rinvsq00         = gmx_mm_inv_pd(rsq00);
 442             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 443             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 444             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 445
 446             /* Load parameters for j particles */
 447             jq0              = _mm_load_sd(charge+jnrA+0);
 448             vdwjidx0A        = 2*vdwtype[jnrA+0];
 449
 450             fjx0             = _mm_setzero_pd();
 451             fjy0             = _mm_setzero_pd();
 452             fjz0             = _mm_setzero_pd();
 453
 454             /**************************
 455              * CALCULATE INTERACTIONS *
 456              **************************/
 457
 458             if (gmx_mm_any_lt(rsq00,rcutoff2))
 459             {
 460
 461             /* Compute parameters for interactions between i and j atoms */
 462             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 463
 464             /* LENNARD-JONES DISPERSION/REPULSION */
 465
 466             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 467             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 468             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 469             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) ,
 470                                           _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 471             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 472
 473             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 474
 475             /* Update potential sum for this i atom from the interaction with this j atom. */
 476             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 477             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 478             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 479
 480             fscal            = fvdw;
 481
 482             fscal            = _mm_and_pd(fscal,cutoff_mask);
 483
 484             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 485
 486             /* Calculate temporary vectorial force */
 487             tx               = _mm_mul_pd(fscal,dx00);
 488             ty               = _mm_mul_pd(fscal,dy00);
 489             tz               = _mm_mul_pd(fscal,dz00);
 490
 491             /* Update vectorial force */
 492             fix0             = _mm_add_pd(fix0,tx);
 493             fiy0             = _mm_add_pd(fiy0,ty);
 494             fiz0             = _mm_add_pd(fiz0,tz);
 495
 496             fjx0             = _mm_add_pd(fjx0,tx);
 497             fjy0             = _mm_add_pd(fjy0,ty);
 498             fjz0             = _mm_add_pd(fjz0,tz);
 499
 500             }
 501
 502             /**************************
 503              * CALCULATE INTERACTIONS *
 504              **************************/
 505
 506             if (gmx_mm_any_lt(rsq10,rcutoff2))
 507             {
 508
 509             /* Compute parameters for interactions between i and j atoms */
 510             qq10             = _mm_mul_pd(iq1,jq0);
 511
 512             /* REACTION-FIELD ELECTROSTATICS */
 513             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 514             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 515
 516             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 517
 518             /* Update potential sum for this i atom from the interaction with this j atom. */
 519             velec            = _mm_and_pd(velec,cutoff_mask);
 520             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 521             velecsum         = _mm_add_pd(velecsum,velec);
 522
 523             fscal            = felec;
 524
 525             fscal            = _mm_and_pd(fscal,cutoff_mask);
 526
 527             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 528
 529             /* Calculate temporary vectorial force */
 530             tx               = _mm_mul_pd(fscal,dx10);
 531             ty               = _mm_mul_pd(fscal,dy10);
 532             tz               = _mm_mul_pd(fscal,dz10);
 533
 534             /* Update vectorial force */
 535             fix1             = _mm_add_pd(fix1,tx);
 536             fiy1             = _mm_add_pd(fiy1,ty);
 537             fiz1             = _mm_add_pd(fiz1,tz);
 538
 539             fjx0             = _mm_add_pd(fjx0,tx);
 540             fjy0             = _mm_add_pd(fjy0,ty);
 541             fjz0             = _mm_add_pd(fjz0,tz);
 542
 543             }
 544
 545             /**************************
 546              * CALCULATE INTERACTIONS *
 547              **************************/
 548
 549             if (gmx_mm_any_lt(rsq20,rcutoff2))
 550             {
 551
 552             /* Compute parameters for interactions between i and j atoms */
 553             qq20             = _mm_mul_pd(iq2,jq0);
 554
 555             /* REACTION-FIELD ELECTROSTATICS */
 556             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 557             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 558
 559             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 560
 561             /* Update potential sum for this i atom from the interaction with this j atom. */
 562             velec            = _mm_and_pd(velec,cutoff_mask);
 563             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 564             velecsum         = _mm_add_pd(velecsum,velec);
 565
 566             fscal            = felec;
 567
 568             fscal            = _mm_and_pd(fscal,cutoff_mask);
 569
 570             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 571
 572             /* Calculate temporary vectorial force */
 573             tx               = _mm_mul_pd(fscal,dx20);
 574             ty               = _mm_mul_pd(fscal,dy20);
 575             tz               = _mm_mul_pd(fscal,dz20);
 576
 577             /* Update vectorial force */
 578             fix2             = _mm_add_pd(fix2,tx);
 579             fiy2             = _mm_add_pd(fiy2,ty);
 580             fiz2             = _mm_add_pd(fiz2,tz);
 581
 582             fjx0             = _mm_add_pd(fjx0,tx);
 583             fjy0             = _mm_add_pd(fjy0,ty);
 584             fjz0             = _mm_add_pd(fjz0,tz);
 585
 586             }
 587
 588             /**************************
 589              * CALCULATE INTERACTIONS *
 590              **************************/
 591
 592             if (gmx_mm_any_lt(rsq30,rcutoff2))
 593             {
 594
 595             /* Compute parameters for interactions between i and j atoms */
 596             qq30             = _mm_mul_pd(iq3,jq0);
 597
 598             /* REACTION-FIELD ELECTROSTATICS */
 599             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_add_pd(rinv30,_mm_mul_pd(krf,rsq30)),crf));
 600             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
 601
 602             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 603
 604             /* Update potential sum for this i atom from the interaction with this j atom. */
 605             velec            = _mm_and_pd(velec,cutoff_mask);
 606             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 607             velecsum         = _mm_add_pd(velecsum,velec);
 608
 609             fscal            = felec;
 610
 611             fscal            = _mm_and_pd(fscal,cutoff_mask);
 612
 613             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 614
 615             /* Calculate temporary vectorial force */
 616             tx               = _mm_mul_pd(fscal,dx30);
 617             ty               = _mm_mul_pd(fscal,dy30);
 618             tz               = _mm_mul_pd(fscal,dz30);
 619
 620             /* Update vectorial force */
 621             fix3             = _mm_add_pd(fix3,tx);
 622             fiy3             = _mm_add_pd(fiy3,ty);
 623             fiz3             = _mm_add_pd(fiz3,tz);
 624
 625             fjx0             = _mm_add_pd(fjx0,tx);
 626             fjy0             = _mm_add_pd(fjy0,ty);
 627             fjz0             = _mm_add_pd(fjz0,tz);
 628
 629             }
 630
 631             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 632
 633             /* Inner loop uses 152 flops */
 634         }
 635
 636         /* End of innermost loop */
 637
 638         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 639                                               f+i_coord_offset,fshift+i_shift_offset);
 640
 641         ggid                        = gid[iidx];
 642         /* Update potential energies */
 643         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 644         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 645
 646         /* Increment number of inner iterations */
 647         inneriter                  += j_index_end - j_index_start;
 648
 649         /* Outer loop uses 26 flops */
 650     }
 651
 652     /* Increment number of outer iterations */
 653     outeriter        += nri;
 654
 655     /* Update outer/inner flops */
 656
 657     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*152);
 658 }
 659 /*
 660  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_F_sse4_1_double
 661  * Electrostatics interaction: ReactionField
 662  * VdW interaction:            LennardJones
 663  * Geometry:                   Water4-Particle
 664  * Calculate force/pot:        Force
 665  */
 666 void
 667 nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_F_sse4_1_double
 668                     (t_nblist                    * gmx_restrict       nlist,
 669                      rvec                        * gmx_restrict          xx,
 670                      rvec                        * gmx_restrict          ff,
 671                      t_forcerec                  * gmx_restrict          fr,
 672                      t_mdatoms                   * gmx_restrict     mdatoms,
 673                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 674                      t_nrnb                      * gmx_restrict        nrnb)
 675 {
 676     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 677      * just 0 for non-waters.
 678      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 679      * jnr indices corresponding to data put in the four positions in the SIMD register.
 680      */
 681     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 682     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 683     int              jnrA,jnrB;
 684     int              j_coord_offsetA,j_coord_offsetB;
 685     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 686     real             rcutoff_scalar;
 687     real             *shiftvec,*fshift,*x,*f;
 688     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 689     int              vdwioffset0;
 690     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 691     int              vdwioffset1;
 692     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 693     int              vdwioffset2;
 694     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 695     int              vdwioffset3;
 696     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 697     int              vdwjidx0A,vdwjidx0B;
 698     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 699     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 700     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 701     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 702     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 703     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 704     real             *charge;
 705     int              nvdwtype;
 706     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 707     int              *vdwtype;
 708     real             *vdwparam;
 709     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 710     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 711     __m128d          dummy_mask,cutoff_mask;
 712     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 713     __m128d          one     = _mm_set1_pd(1.0);
 714     __m128d          two     = _mm_set1_pd(2.0);
 715     x                = xx[0];
 716     f                = ff[0];
 717
 718     nri              = nlist->nri;
 719     iinr             = nlist->iinr;
 720     jindex           = nlist->jindex;
 721     jjnr             = nlist->jjnr;
 722     shiftidx         = nlist->shift;
 723     gid              = nlist->gid;
 724     shiftvec         = fr->shift_vec[0];
 725     fshift           = fr->fshift[0];
 726     facel            = _mm_set1_pd(fr->epsfac);
 727     charge           = mdatoms->chargeA;
 728     krf              = _mm_set1_pd(fr->ic->k_rf);
 729     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 730     crf              = _mm_set1_pd(fr->ic->c_rf);
 731     nvdwtype         = fr->ntype;
 732     vdwparam         = fr->nbfp;
 733     vdwtype          = mdatoms->typeA;
 734
 735     /* Setup water-specific parameters */
 736     inr              = nlist->iinr[0];
 737     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 738     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 739     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 740     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 741
 742     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 743     rcutoff_scalar   = fr->rcoulomb;
 744     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 745     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 746
 747     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 748     rvdw             = _mm_set1_pd(fr->rvdw);
 749
 750     /* Avoid stupid compiler warnings */
 751     jnrA = jnrB = 0;
 752     j_coord_offsetA = 0;
 753     j_coord_offsetB = 0;
 754
 755     outeriter        = 0;
 756     inneriter        = 0;
 757
 758     /* Start outer loop over neighborlists */
 759     for(iidx=0; iidx<nri; iidx++)
 760     {
 761         /* Load shift vector for this list */
 762         i_shift_offset   = DIM*shiftidx[iidx];
 763
 764         /* Load limits for loop over neighbors */
 765         j_index_start    = jindex[iidx];
 766         j_index_end      = jindex[iidx+1];
 767
 768         /* Get outer coordinate index */
 769         inr              = iinr[iidx];
 770         i_coord_offset   = DIM*inr;
 771
 772         /* Load i particle coords and add shift vector */
 773         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 774                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 775
 776         fix0             = _mm_setzero_pd();
 777         fiy0             = _mm_setzero_pd();
 778         fiz0             = _mm_setzero_pd();
 779         fix1             = _mm_setzero_pd();
 780         fiy1             = _mm_setzero_pd();
 781         fiz1             = _mm_setzero_pd();
 782         fix2             = _mm_setzero_pd();
 783         fiy2             = _mm_setzero_pd();
 784         fiz2             = _mm_setzero_pd();
 785         fix3             = _mm_setzero_pd();
 786         fiy3             = _mm_setzero_pd();
 787         fiz3             = _mm_setzero_pd();
 788
 789         /* Start inner kernel loop */
 790         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 791         {
 792
 793             /* Get j neighbor index, and coordinate index */
 794             jnrA             = jjnr[jidx];
 795             jnrB             = jjnr[jidx+1];
 796             j_coord_offsetA  = DIM*jnrA;
 797             j_coord_offsetB  = DIM*jnrB;
 798
 799             /* load j atom coordinates */
 800             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 801                                               &jx0,&jy0,&jz0);
 802
 803             /* Calculate displacement vector */
 804             dx00             = _mm_sub_pd(ix0,jx0);
 805             dy00             = _mm_sub_pd(iy0,jy0);
 806             dz00             = _mm_sub_pd(iz0,jz0);
 807             dx10             = _mm_sub_pd(ix1,jx0);
 808             dy10             = _mm_sub_pd(iy1,jy0);
 809             dz10             = _mm_sub_pd(iz1,jz0);
 810             dx20             = _mm_sub_pd(ix2,jx0);
 811             dy20             = _mm_sub_pd(iy2,jy0);
 812             dz20             = _mm_sub_pd(iz2,jz0);
 813             dx30             = _mm_sub_pd(ix3,jx0);
 814             dy30             = _mm_sub_pd(iy3,jy0);
 815             dz30             = _mm_sub_pd(iz3,jz0);
 816
 817             /* Calculate squared distance and things based on it */
 818             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 819             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 820             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 821             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 822
 823             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 824             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 825             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 826
 827             rinvsq00         = gmx_mm_inv_pd(rsq00);
 828             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 829             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 830             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 831
 832             /* Load parameters for j particles */
 833             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 834             vdwjidx0A        = 2*vdwtype[jnrA+0];
 835             vdwjidx0B        = 2*vdwtype[jnrB+0];
 836
 837             fjx0             = _mm_setzero_pd();
 838             fjy0             = _mm_setzero_pd();
 839             fjz0             = _mm_setzero_pd();
 840
 841             /**************************
 842              * CALCULATE INTERACTIONS *
 843              **************************/
 844
 845             if (gmx_mm_any_lt(rsq00,rcutoff2))
 846             {
 847
 848             /* Compute parameters for interactions between i and j atoms */
 849             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 850                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 851
 852             /* LENNARD-JONES DISPERSION/REPULSION */
 853
 854             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 855             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 856
 857             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 858
 859             fscal            = fvdw;
 860
 861             fscal            = _mm_and_pd(fscal,cutoff_mask);
 862
 863             /* Calculate temporary vectorial force */
 864             tx               = _mm_mul_pd(fscal,dx00);
 865             ty               = _mm_mul_pd(fscal,dy00);
 866             tz               = _mm_mul_pd(fscal,dz00);
 867
 868             /* Update vectorial force */
 869             fix0             = _mm_add_pd(fix0,tx);
 870             fiy0             = _mm_add_pd(fiy0,ty);
 871             fiz0             = _mm_add_pd(fiz0,tz);
 872
 873             fjx0             = _mm_add_pd(fjx0,tx);
 874             fjy0             = _mm_add_pd(fjy0,ty);
 875             fjz0             = _mm_add_pd(fjz0,tz);
 876
 877             }
 878
 879             /**************************
 880              * CALCULATE INTERACTIONS *
 881              **************************/
 882
 883             if (gmx_mm_any_lt(rsq10,rcutoff2))
 884             {
 885
 886             /* Compute parameters for interactions between i and j atoms */
 887             qq10             = _mm_mul_pd(iq1,jq0);
 888
 889             /* REACTION-FIELD ELECTROSTATICS */
 890             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 891
 892             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 893
 894             fscal            = felec;
 895
 896             fscal            = _mm_and_pd(fscal,cutoff_mask);
 897
 898             /* Calculate temporary vectorial force */
 899             tx               = _mm_mul_pd(fscal,dx10);
 900             ty               = _mm_mul_pd(fscal,dy10);
 901             tz               = _mm_mul_pd(fscal,dz10);
 902
 903             /* Update vectorial force */
 904             fix1             = _mm_add_pd(fix1,tx);
 905             fiy1             = _mm_add_pd(fiy1,ty);
 906             fiz1             = _mm_add_pd(fiz1,tz);
 907
 908             fjx0             = _mm_add_pd(fjx0,tx);
 909             fjy0             = _mm_add_pd(fjy0,ty);
 910             fjz0             = _mm_add_pd(fjz0,tz);
 911
 912             }
 913
 914             /**************************
 915              * CALCULATE INTERACTIONS *
 916              **************************/
 917
 918             if (gmx_mm_any_lt(rsq20,rcutoff2))
 919             {
 920
 921             /* Compute parameters for interactions between i and j atoms */
 922             qq20             = _mm_mul_pd(iq2,jq0);
 923
 924             /* REACTION-FIELD ELECTROSTATICS */
 925             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 926
 927             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 928
 929             fscal            = felec;
 930
 931             fscal            = _mm_and_pd(fscal,cutoff_mask);
 932
 933             /* Calculate temporary vectorial force */
 934             tx               = _mm_mul_pd(fscal,dx20);
 935             ty               = _mm_mul_pd(fscal,dy20);
 936             tz               = _mm_mul_pd(fscal,dz20);
 937
 938             /* Update vectorial force */
 939             fix2             = _mm_add_pd(fix2,tx);
 940             fiy2             = _mm_add_pd(fiy2,ty);
 941             fiz2             = _mm_add_pd(fiz2,tz);
 942
 943             fjx0             = _mm_add_pd(fjx0,tx);
 944             fjy0             = _mm_add_pd(fjy0,ty);
 945             fjz0             = _mm_add_pd(fjz0,tz);
 946
 947             }
 948
 949             /**************************
 950              * CALCULATE INTERACTIONS *
 951              **************************/
 952
 953             if (gmx_mm_any_lt(rsq30,rcutoff2))
 954             {
 955
 956             /* Compute parameters for interactions between i and j atoms */
 957             qq30             = _mm_mul_pd(iq3,jq0);
 958
 959             /* REACTION-FIELD ELECTROSTATICS */
 960             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
 961
 962             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 963
 964             fscal            = felec;
 965
 966             fscal            = _mm_and_pd(fscal,cutoff_mask);
 967
 968             /* Calculate temporary vectorial force */
 969             tx               = _mm_mul_pd(fscal,dx30);
 970             ty               = _mm_mul_pd(fscal,dy30);
 971             tz               = _mm_mul_pd(fscal,dz30);
 972
 973             /* Update vectorial force */
 974             fix3             = _mm_add_pd(fix3,tx);
 975             fiy3             = _mm_add_pd(fiy3,ty);
 976             fiz3             = _mm_add_pd(fiz3,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             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 985
 986             /* Inner loop uses 123 flops */
 987         }
 988
 989         if(jidx<j_index_end)
 990         {
 991
 992             jnrA             = jjnr[jidx];
 993             j_coord_offsetA  = DIM*jnrA;
 994
 995             /* load j atom coordinates */
 996             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 997                                               &jx0,&jy0,&jz0);
 998
 999             /* Calculate displacement vector */
1000             dx00             = _mm_sub_pd(ix0,jx0);
1001             dy00             = _mm_sub_pd(iy0,jy0);
1002             dz00             = _mm_sub_pd(iz0,jz0);
1003             dx10             = _mm_sub_pd(ix1,jx0);
1004             dy10             = _mm_sub_pd(iy1,jy0);
1005             dz10             = _mm_sub_pd(iz1,jz0);
1006             dx20             = _mm_sub_pd(ix2,jx0);
1007             dy20             = _mm_sub_pd(iy2,jy0);
1008             dz20             = _mm_sub_pd(iz2,jz0);
1009             dx30             = _mm_sub_pd(ix3,jx0);
1010             dy30             = _mm_sub_pd(iy3,jy0);
1011             dz30             = _mm_sub_pd(iz3,jz0);
1012
1013             /* Calculate squared distance and things based on it */
1014             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1015             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1016             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1017             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1018
1019             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1020             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1021             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1022
1023             rinvsq00         = gmx_mm_inv_pd(rsq00);
1024             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1025             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1026             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1027
1028             /* Load parameters for j particles */
1029             jq0              = _mm_load_sd(charge+jnrA+0);
1030             vdwjidx0A        = 2*vdwtype[jnrA+0];
1031
1032             fjx0             = _mm_setzero_pd();
1033             fjy0             = _mm_setzero_pd();
1034             fjz0             = _mm_setzero_pd();
1035
1036             /**************************
1037              * CALCULATE INTERACTIONS *
1038              **************************/
1039
1040             if (gmx_mm_any_lt(rsq00,rcutoff2))
1041             {
1042
1043             /* Compute parameters for interactions between i and j atoms */
1044             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1045
1046             /* LENNARD-JONES DISPERSION/REPULSION */
1047
1048             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1049             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1050
1051             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1052
1053             fscal            = fvdw;
1054
1055             fscal            = _mm_and_pd(fscal,cutoff_mask);
1056
1057             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1058
1059             /* Calculate temporary vectorial force */
1060             tx               = _mm_mul_pd(fscal,dx00);
1061             ty               = _mm_mul_pd(fscal,dy00);
1062             tz               = _mm_mul_pd(fscal,dz00);
1063
1064             /* Update vectorial force */
1065             fix0             = _mm_add_pd(fix0,tx);
1066             fiy0             = _mm_add_pd(fiy0,ty);
1067             fiz0             = _mm_add_pd(fiz0,tz);
1068
1069             fjx0             = _mm_add_pd(fjx0,tx);
1070             fjy0             = _mm_add_pd(fjy0,ty);
1071             fjz0             = _mm_add_pd(fjz0,tz);
1072
1073             }
1074
1075             /**************************
1076              * CALCULATE INTERACTIONS *
1077              **************************/
1078
1079             if (gmx_mm_any_lt(rsq10,rcutoff2))
1080             {
1081
1082             /* Compute parameters for interactions between i and j atoms */
1083             qq10             = _mm_mul_pd(iq1,jq0);
1084
1085             /* REACTION-FIELD ELECTROSTATICS */
1086             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
1087
1088             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1089
1090             fscal            = felec;
1091
1092             fscal            = _mm_and_pd(fscal,cutoff_mask);
1093
1094             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1095
1096             /* Calculate temporary vectorial force */
1097             tx               = _mm_mul_pd(fscal,dx10);
1098             ty               = _mm_mul_pd(fscal,dy10);
1099             tz               = _mm_mul_pd(fscal,dz10);
1100
1101             /* Update vectorial force */
1102             fix1             = _mm_add_pd(fix1,tx);
1103             fiy1             = _mm_add_pd(fiy1,ty);
1104             fiz1             = _mm_add_pd(fiz1,tz);
1105
1106             fjx0             = _mm_add_pd(fjx0,tx);
1107             fjy0             = _mm_add_pd(fjy0,ty);
1108             fjz0             = _mm_add_pd(fjz0,tz);
1109
1110             }
1111
1112             /**************************
1113              * CALCULATE INTERACTIONS *
1114              **************************/
1115
1116             if (gmx_mm_any_lt(rsq20,rcutoff2))
1117             {
1118
1119             /* Compute parameters for interactions between i and j atoms */
1120             qq20             = _mm_mul_pd(iq2,jq0);
1121
1122             /* REACTION-FIELD ELECTROSTATICS */
1123             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
1124
1125             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1126
1127             fscal            = felec;
1128
1129             fscal            = _mm_and_pd(fscal,cutoff_mask);
1130
1131             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1132
1133             /* Calculate temporary vectorial force */
1134             tx               = _mm_mul_pd(fscal,dx20);
1135             ty               = _mm_mul_pd(fscal,dy20);
1136             tz               = _mm_mul_pd(fscal,dz20);
1137
1138             /* Update vectorial force */
1139             fix2             = _mm_add_pd(fix2,tx);
1140             fiy2             = _mm_add_pd(fiy2,ty);
1141             fiz2             = _mm_add_pd(fiz2,tz);
1142
1143             fjx0             = _mm_add_pd(fjx0,tx);
1144             fjy0             = _mm_add_pd(fjy0,ty);
1145             fjz0             = _mm_add_pd(fjz0,tz);
1146
1147             }
1148
1149             /**************************
1150              * CALCULATE INTERACTIONS *
1151              **************************/
1152
1153             if (gmx_mm_any_lt(rsq30,rcutoff2))
1154             {
1155
1156             /* Compute parameters for interactions between i and j atoms */
1157             qq30             = _mm_mul_pd(iq3,jq0);
1158
1159             /* REACTION-FIELD ELECTROSTATICS */
1160             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
1161
1162             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1163
1164             fscal            = felec;
1165
1166             fscal            = _mm_and_pd(fscal,cutoff_mask);
1167
1168             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1169
1170             /* Calculate temporary vectorial force */
1171             tx               = _mm_mul_pd(fscal,dx30);
1172             ty               = _mm_mul_pd(fscal,dy30);
1173             tz               = _mm_mul_pd(fscal,dz30);
1174
1175             /* Update vectorial force */
1176             fix3             = _mm_add_pd(fix3,tx);
1177             fiy3             = _mm_add_pd(fiy3,ty);
1178             fiz3             = _mm_add_pd(fiz3,tz);
1179
1180             fjx0             = _mm_add_pd(fjx0,tx);
1181             fjy0             = _mm_add_pd(fjy0,ty);
1182             fjz0             = _mm_add_pd(fjz0,tz);
1183
1184             }
1185
1186             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1187
1188             /* Inner loop uses 123 flops */
1189         }
1190
1191         /* End of innermost loop */
1192
1193         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1194                                               f+i_coord_offset,fshift+i_shift_offset);
1195
1196         /* Increment number of inner iterations */
1197         inneriter                  += j_index_end - j_index_start;
1198
1199         /* Outer loop uses 24 flops */
1200     }
1201
1202     /* Increment number of outer iterations */
1203     outeriter        += nri;
1204
1205     /* Update outer/inner flops */
1206
1207     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*123);
1208 }