src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecCoul_VdwNone_GeomW3P1_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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse4_1_double.h"
  50 #include "kernelutil_x86_sse4_1_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomW3P1_VF_sse4_1_double
  54  * Electrostatics interaction: Coulomb
  55  * VdW interaction:            None
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCoul_VdwNone_GeomW3P1_VF_sse4_1_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  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          velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     __m128d          dummy_mask,cutoff_mask;
  96     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  97     __m128d          one     = _mm_set1_pd(1.0);
  98     __m128d          two     = _mm_set1_pd(2.0);
  99     x                = xx[0];
 100     f                = ff[0];
 101
 102     nri              = nlist->nri;
 103     iinr             = nlist->iinr;
 104     jindex           = nlist->jindex;
 105     jjnr             = nlist->jjnr;
 106     shiftidx         = nlist->shift;
 107     gid              = nlist->gid;
 108     shiftvec         = fr->shift_vec[0];
 109     fshift           = fr->fshift[0];
 110     facel            = _mm_set1_pd(fr->epsfac);
 111     charge           = mdatoms->chargeA;
 112
 113     /* Setup water-specific parameters */
 114     inr              = nlist->iinr[0];
 115     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 116     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 117     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 118
 119     /* Avoid stupid compiler warnings */
 120     jnrA = jnrB = 0;
 121     j_coord_offsetA = 0;
 122     j_coord_offsetB = 0;
 123
 124     outeriter        = 0;
 125     inneriter        = 0;
 126
 127     /* Start outer loop over neighborlists */
 128     for(iidx=0; iidx<nri; iidx++)
 129     {
 130         /* Load shift vector for this list */
 131         i_shift_offset   = DIM*shiftidx[iidx];
 132
 133         /* Load limits for loop over neighbors */
 134         j_index_start    = jindex[iidx];
 135         j_index_end      = jindex[iidx+1];
 136
 137         /* Get outer coordinate index */
 138         inr              = iinr[iidx];
 139         i_coord_offset   = DIM*inr;
 140
 141         /* Load i particle coords and add shift vector */
 142         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 143                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 144
 145         fix0             = _mm_setzero_pd();
 146         fiy0             = _mm_setzero_pd();
 147         fiz0             = _mm_setzero_pd();
 148         fix1             = _mm_setzero_pd();
 149         fiy1             = _mm_setzero_pd();
 150         fiz1             = _mm_setzero_pd();
 151         fix2             = _mm_setzero_pd();
 152         fiy2             = _mm_setzero_pd();
 153         fiz2             = _mm_setzero_pd();
 154
 155         /* Reset potential sums */
 156         velecsum         = _mm_setzero_pd();
 157
 158         /* Start inner kernel loop */
 159         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 160         {
 161
 162             /* Get j neighbor index, and coordinate index */
 163             jnrA             = jjnr[jidx];
 164             jnrB             = jjnr[jidx+1];
 165             j_coord_offsetA  = DIM*jnrA;
 166             j_coord_offsetB  = DIM*jnrB;
 167
 168             /* load j atom coordinates */
 169             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 170                                               &jx0,&jy0,&jz0);
 171
 172             /* Calculate displacement vector */
 173             dx00             = _mm_sub_pd(ix0,jx0);
 174             dy00             = _mm_sub_pd(iy0,jy0);
 175             dz00             = _mm_sub_pd(iz0,jz0);
 176             dx10             = _mm_sub_pd(ix1,jx0);
 177             dy10             = _mm_sub_pd(iy1,jy0);
 178             dz10             = _mm_sub_pd(iz1,jz0);
 179             dx20             = _mm_sub_pd(ix2,jx0);
 180             dy20             = _mm_sub_pd(iy2,jy0);
 181             dz20             = _mm_sub_pd(iz2,jz0);
 182
 183             /* Calculate squared distance and things based on it */
 184             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 185             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 186             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 187
 188             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 189             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 190             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 191
 192             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 193             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 194             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 195
 196             /* Load parameters for j particles */
 197             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 198
 199             fjx0             = _mm_setzero_pd();
 200             fjy0             = _mm_setzero_pd();
 201             fjz0             = _mm_setzero_pd();
 202
 203             /**************************
 204              * CALCULATE INTERACTIONS *
 205              **************************/
 206
 207             /* Compute parameters for interactions between i and j atoms */
 208             qq00             = _mm_mul_pd(iq0,jq0);
 209
 210             /* COULOMB ELECTROSTATICS */
 211             velec            = _mm_mul_pd(qq00,rinv00);
 212             felec            = _mm_mul_pd(velec,rinvsq00);
 213
 214             /* Update potential sum for this i atom from the interaction with this j atom. */
 215             velecsum         = _mm_add_pd(velecsum,velec);
 216
 217             fscal            = felec;
 218
 219             /* Calculate temporary vectorial force */
 220             tx               = _mm_mul_pd(fscal,dx00);
 221             ty               = _mm_mul_pd(fscal,dy00);
 222             tz               = _mm_mul_pd(fscal,dz00);
 223
 224             /* Update vectorial force */
 225             fix0             = _mm_add_pd(fix0,tx);
 226             fiy0             = _mm_add_pd(fiy0,ty);
 227             fiz0             = _mm_add_pd(fiz0,tz);
 228
 229             fjx0             = _mm_add_pd(fjx0,tx);
 230             fjy0             = _mm_add_pd(fjy0,ty);
 231             fjz0             = _mm_add_pd(fjz0,tz);
 232
 233             /**************************
 234              * CALCULATE INTERACTIONS *
 235              **************************/
 236
 237             /* Compute parameters for interactions between i and j atoms */
 238             qq10             = _mm_mul_pd(iq1,jq0);
 239
 240             /* COULOMB ELECTROSTATICS */
 241             velec            = _mm_mul_pd(qq10,rinv10);
 242             felec            = _mm_mul_pd(velec,rinvsq10);
 243
 244             /* Update potential sum for this i atom from the interaction with this j atom. */
 245             velecsum         = _mm_add_pd(velecsum,velec);
 246
 247             fscal            = felec;
 248
 249             /* Calculate temporary vectorial force */
 250             tx               = _mm_mul_pd(fscal,dx10);
 251             ty               = _mm_mul_pd(fscal,dy10);
 252             tz               = _mm_mul_pd(fscal,dz10);
 253
 254             /* Update vectorial force */
 255             fix1             = _mm_add_pd(fix1,tx);
 256             fiy1             = _mm_add_pd(fiy1,ty);
 257             fiz1             = _mm_add_pd(fiz1,tz);
 258
 259             fjx0             = _mm_add_pd(fjx0,tx);
 260             fjy0             = _mm_add_pd(fjy0,ty);
 261             fjz0             = _mm_add_pd(fjz0,tz);
 262
 263             /**************************
 264              * CALCULATE INTERACTIONS *
 265              **************************/
 266
 267             /* Compute parameters for interactions between i and j atoms */
 268             qq20             = _mm_mul_pd(iq2,jq0);
 269
 270             /* COULOMB ELECTROSTATICS */
 271             velec            = _mm_mul_pd(qq20,rinv20);
 272             felec            = _mm_mul_pd(velec,rinvsq20);
 273
 274             /* Update potential sum for this i atom from the interaction with this j atom. */
 275             velecsum         = _mm_add_pd(velecsum,velec);
 276
 277             fscal            = felec;
 278
 279             /* Calculate temporary vectorial force */
 280             tx               = _mm_mul_pd(fscal,dx20);
 281             ty               = _mm_mul_pd(fscal,dy20);
 282             tz               = _mm_mul_pd(fscal,dz20);
 283
 284             /* Update vectorial force */
 285             fix2             = _mm_add_pd(fix2,tx);
 286             fiy2             = _mm_add_pd(fiy2,ty);
 287             fiz2             = _mm_add_pd(fiz2,tz);
 288
 289             fjx0             = _mm_add_pd(fjx0,tx);
 290             fjy0             = _mm_add_pd(fjy0,ty);
 291             fjz0             = _mm_add_pd(fjz0,tz);
 292
 293             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 294
 295             /* Inner loop uses 87 flops */
 296         }
 297
 298         if(jidx<j_index_end)
 299         {
 300
 301             jnrA             = jjnr[jidx];
 302             j_coord_offsetA  = DIM*jnrA;
 303
 304             /* load j atom coordinates */
 305             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 306                                               &jx0,&jy0,&jz0);
 307
 308             /* Calculate displacement vector */
 309             dx00             = _mm_sub_pd(ix0,jx0);
 310             dy00             = _mm_sub_pd(iy0,jy0);
 311             dz00             = _mm_sub_pd(iz0,jz0);
 312             dx10             = _mm_sub_pd(ix1,jx0);
 313             dy10             = _mm_sub_pd(iy1,jy0);
 314             dz10             = _mm_sub_pd(iz1,jz0);
 315             dx20             = _mm_sub_pd(ix2,jx0);
 316             dy20             = _mm_sub_pd(iy2,jy0);
 317             dz20             = _mm_sub_pd(iz2,jz0);
 318
 319             /* Calculate squared distance and things based on it */
 320             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 321             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 322             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 323
 324             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 325             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 326             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 327
 328             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 329             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 330             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 331
 332             /* Load parameters for j particles */
 333             jq0              = _mm_load_sd(charge+jnrA+0);
 334
 335             fjx0             = _mm_setzero_pd();
 336             fjy0             = _mm_setzero_pd();
 337             fjz0             = _mm_setzero_pd();
 338
 339             /**************************
 340              * CALCULATE INTERACTIONS *
 341              **************************/
 342
 343             /* Compute parameters for interactions between i and j atoms */
 344             qq00             = _mm_mul_pd(iq0,jq0);
 345
 346             /* COULOMB ELECTROSTATICS */
 347             velec            = _mm_mul_pd(qq00,rinv00);
 348             felec            = _mm_mul_pd(velec,rinvsq00);
 349
 350             /* Update potential sum for this i atom from the interaction with this j atom. */
 351             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 352             velecsum         = _mm_add_pd(velecsum,velec);
 353
 354             fscal            = felec;
 355
 356             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 357
 358             /* Calculate temporary vectorial force */
 359             tx               = _mm_mul_pd(fscal,dx00);
 360             ty               = _mm_mul_pd(fscal,dy00);
 361             tz               = _mm_mul_pd(fscal,dz00);
 362
 363             /* Update vectorial force */
 364             fix0             = _mm_add_pd(fix0,tx);
 365             fiy0             = _mm_add_pd(fiy0,ty);
 366             fiz0             = _mm_add_pd(fiz0,tz);
 367
 368             fjx0             = _mm_add_pd(fjx0,tx);
 369             fjy0             = _mm_add_pd(fjy0,ty);
 370             fjz0             = _mm_add_pd(fjz0,tz);
 371
 372             /**************************
 373              * CALCULATE INTERACTIONS *
 374              **************************/
 375
 376             /* Compute parameters for interactions between i and j atoms */
 377             qq10             = _mm_mul_pd(iq1,jq0);
 378
 379             /* COULOMB ELECTROSTATICS */
 380             velec            = _mm_mul_pd(qq10,rinv10);
 381             felec            = _mm_mul_pd(velec,rinvsq10);
 382
 383             /* Update potential sum for this i atom from the interaction with this j atom. */
 384             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 385             velecsum         = _mm_add_pd(velecsum,velec);
 386
 387             fscal            = felec;
 388
 389             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 390
 391             /* Calculate temporary vectorial force */
 392             tx               = _mm_mul_pd(fscal,dx10);
 393             ty               = _mm_mul_pd(fscal,dy10);
 394             tz               = _mm_mul_pd(fscal,dz10);
 395
 396             /* Update vectorial force */
 397             fix1             = _mm_add_pd(fix1,tx);
 398             fiy1             = _mm_add_pd(fiy1,ty);
 399             fiz1             = _mm_add_pd(fiz1,tz);
 400
 401             fjx0             = _mm_add_pd(fjx0,tx);
 402             fjy0             = _mm_add_pd(fjy0,ty);
 403             fjz0             = _mm_add_pd(fjz0,tz);
 404
 405             /**************************
 406              * CALCULATE INTERACTIONS *
 407              **************************/
 408
 409             /* Compute parameters for interactions between i and j atoms */
 410             qq20             = _mm_mul_pd(iq2,jq0);
 411
 412             /* COULOMB ELECTROSTATICS */
 413             velec            = _mm_mul_pd(qq20,rinv20);
 414             felec            = _mm_mul_pd(velec,rinvsq20);
 415
 416             /* Update potential sum for this i atom from the interaction with this j atom. */
 417             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 418             velecsum         = _mm_add_pd(velecsum,velec);
 419
 420             fscal            = felec;
 421
 422             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 423
 424             /* Calculate temporary vectorial force */
 425             tx               = _mm_mul_pd(fscal,dx20);
 426             ty               = _mm_mul_pd(fscal,dy20);
 427             tz               = _mm_mul_pd(fscal,dz20);
 428
 429             /* Update vectorial force */
 430             fix2             = _mm_add_pd(fix2,tx);
 431             fiy2             = _mm_add_pd(fiy2,ty);
 432             fiz2             = _mm_add_pd(fiz2,tz);
 433
 434             fjx0             = _mm_add_pd(fjx0,tx);
 435             fjy0             = _mm_add_pd(fjy0,ty);
 436             fjz0             = _mm_add_pd(fjz0,tz);
 437
 438             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 439
 440             /* Inner loop uses 87 flops */
 441         }
 442
 443         /* End of innermost loop */
 444
 445         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 446                                               f+i_coord_offset,fshift+i_shift_offset);
 447
 448         ggid                        = gid[iidx];
 449         /* Update potential energies */
 450         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 451
 452         /* Increment number of inner iterations */
 453         inneriter                  += j_index_end - j_index_start;
 454
 455         /* Outer loop uses 19 flops */
 456     }
 457
 458     /* Increment number of outer iterations */
 459     outeriter        += nri;
 460
 461     /* Update outer/inner flops */
 462
 463     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*87);
 464 }
 465 /*
 466  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomW3P1_F_sse4_1_double
 467  * Electrostatics interaction: Coulomb
 468  * VdW interaction:            None
 469  * Geometry:                   Water3-Particle
 470  * Calculate force/pot:        Force
 471  */
 472 void
 473 nb_kernel_ElecCoul_VdwNone_GeomW3P1_F_sse4_1_double
 474                     (t_nblist                    * gmx_restrict       nlist,
 475                      rvec                        * gmx_restrict          xx,
 476                      rvec                        * gmx_restrict          ff,
 477                      t_forcerec                  * gmx_restrict          fr,
 478                      t_mdatoms                   * gmx_restrict     mdatoms,
 479                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 480                      t_nrnb                      * gmx_restrict        nrnb)
 481 {
 482     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 483      * just 0 for non-waters.
 484      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 485      * jnr indices corresponding to data put in the four positions in the SIMD register.
 486      */
 487     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 488     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 489     int              jnrA,jnrB;
 490     int              j_coord_offsetA,j_coord_offsetB;
 491     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 492     real             rcutoff_scalar;
 493     real             *shiftvec,*fshift,*x,*f;
 494     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 495     int              vdwioffset0;
 496     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 497     int              vdwioffset1;
 498     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 499     int              vdwioffset2;
 500     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 501     int              vdwjidx0A,vdwjidx0B;
 502     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 503     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 504     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 505     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 506     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 507     real             *charge;
 508     __m128d          dummy_mask,cutoff_mask;
 509     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 510     __m128d          one     = _mm_set1_pd(1.0);
 511     __m128d          two     = _mm_set1_pd(2.0);
 512     x                = xx[0];
 513     f                = ff[0];
 514
 515     nri              = nlist->nri;
 516     iinr             = nlist->iinr;
 517     jindex           = nlist->jindex;
 518     jjnr             = nlist->jjnr;
 519     shiftidx         = nlist->shift;
 520     gid              = nlist->gid;
 521     shiftvec         = fr->shift_vec[0];
 522     fshift           = fr->fshift[0];
 523     facel            = _mm_set1_pd(fr->epsfac);
 524     charge           = mdatoms->chargeA;
 525
 526     /* Setup water-specific parameters */
 527     inr              = nlist->iinr[0];
 528     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 529     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 530     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 531
 532     /* Avoid stupid compiler warnings */
 533     jnrA = jnrB = 0;
 534     j_coord_offsetA = 0;
 535     j_coord_offsetB = 0;
 536
 537     outeriter        = 0;
 538     inneriter        = 0;
 539
 540     /* Start outer loop over neighborlists */
 541     for(iidx=0; iidx<nri; iidx++)
 542     {
 543         /* Load shift vector for this list */
 544         i_shift_offset   = DIM*shiftidx[iidx];
 545
 546         /* Load limits for loop over neighbors */
 547         j_index_start    = jindex[iidx];
 548         j_index_end      = jindex[iidx+1];
 549
 550         /* Get outer coordinate index */
 551         inr              = iinr[iidx];
 552         i_coord_offset   = DIM*inr;
 553
 554         /* Load i particle coords and add shift vector */
 555         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 556                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 557
 558         fix0             = _mm_setzero_pd();
 559         fiy0             = _mm_setzero_pd();
 560         fiz0             = _mm_setzero_pd();
 561         fix1             = _mm_setzero_pd();
 562         fiy1             = _mm_setzero_pd();
 563         fiz1             = _mm_setzero_pd();
 564         fix2             = _mm_setzero_pd();
 565         fiy2             = _mm_setzero_pd();
 566         fiz2             = _mm_setzero_pd();
 567
 568         /* Start inner kernel loop */
 569         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 570         {
 571
 572             /* Get j neighbor index, and coordinate index */
 573             jnrA             = jjnr[jidx];
 574             jnrB             = jjnr[jidx+1];
 575             j_coord_offsetA  = DIM*jnrA;
 576             j_coord_offsetB  = DIM*jnrB;
 577
 578             /* load j atom coordinates */
 579             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 580                                               &jx0,&jy0,&jz0);
 581
 582             /* Calculate displacement vector */
 583             dx00             = _mm_sub_pd(ix0,jx0);
 584             dy00             = _mm_sub_pd(iy0,jy0);
 585             dz00             = _mm_sub_pd(iz0,jz0);
 586             dx10             = _mm_sub_pd(ix1,jx0);
 587             dy10             = _mm_sub_pd(iy1,jy0);
 588             dz10             = _mm_sub_pd(iz1,jz0);
 589             dx20             = _mm_sub_pd(ix2,jx0);
 590             dy20             = _mm_sub_pd(iy2,jy0);
 591             dz20             = _mm_sub_pd(iz2,jz0);
 592
 593             /* Calculate squared distance and things based on it */
 594             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 595             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 596             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 597
 598             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 599             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 600             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 601
 602             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 603             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 604             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 605
 606             /* Load parameters for j particles */
 607             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 608
 609             fjx0             = _mm_setzero_pd();
 610             fjy0             = _mm_setzero_pd();
 611             fjz0             = _mm_setzero_pd();
 612
 613             /**************************
 614              * CALCULATE INTERACTIONS *
 615              **************************/
 616
 617             /* Compute parameters for interactions between i and j atoms */
 618             qq00             = _mm_mul_pd(iq0,jq0);
 619
 620             /* COULOMB ELECTROSTATICS */
 621             velec            = _mm_mul_pd(qq00,rinv00);
 622             felec            = _mm_mul_pd(velec,rinvsq00);
 623
 624             fscal            = felec;
 625
 626             /* Calculate temporary vectorial force */
 627             tx               = _mm_mul_pd(fscal,dx00);
 628             ty               = _mm_mul_pd(fscal,dy00);
 629             tz               = _mm_mul_pd(fscal,dz00);
 630
 631             /* Update vectorial force */
 632             fix0             = _mm_add_pd(fix0,tx);
 633             fiy0             = _mm_add_pd(fiy0,ty);
 634             fiz0             = _mm_add_pd(fiz0,tz);
 635
 636             fjx0             = _mm_add_pd(fjx0,tx);
 637             fjy0             = _mm_add_pd(fjy0,ty);
 638             fjz0             = _mm_add_pd(fjz0,tz);
 639
 640             /**************************
 641              * CALCULATE INTERACTIONS *
 642              **************************/
 643
 644             /* Compute parameters for interactions between i and j atoms */
 645             qq10             = _mm_mul_pd(iq1,jq0);
 646
 647             /* COULOMB ELECTROSTATICS */
 648             velec            = _mm_mul_pd(qq10,rinv10);
 649             felec            = _mm_mul_pd(velec,rinvsq10);
 650
 651             fscal            = felec;
 652
 653             /* Calculate temporary vectorial force */
 654             tx               = _mm_mul_pd(fscal,dx10);
 655             ty               = _mm_mul_pd(fscal,dy10);
 656             tz               = _mm_mul_pd(fscal,dz10);
 657
 658             /* Update vectorial force */
 659             fix1             = _mm_add_pd(fix1,tx);
 660             fiy1             = _mm_add_pd(fiy1,ty);
 661             fiz1             = _mm_add_pd(fiz1,tz);
 662
 663             fjx0             = _mm_add_pd(fjx0,tx);
 664             fjy0             = _mm_add_pd(fjy0,ty);
 665             fjz0             = _mm_add_pd(fjz0,tz);
 666
 667             /**************************
 668              * CALCULATE INTERACTIONS *
 669              **************************/
 670
 671             /* Compute parameters for interactions between i and j atoms */
 672             qq20             = _mm_mul_pd(iq2,jq0);
 673
 674             /* COULOMB ELECTROSTATICS */
 675             velec            = _mm_mul_pd(qq20,rinv20);
 676             felec            = _mm_mul_pd(velec,rinvsq20);
 677
 678             fscal            = felec;
 679
 680             /* Calculate temporary vectorial force */
 681             tx               = _mm_mul_pd(fscal,dx20);
 682             ty               = _mm_mul_pd(fscal,dy20);
 683             tz               = _mm_mul_pd(fscal,dz20);
 684
 685             /* Update vectorial force */
 686             fix2             = _mm_add_pd(fix2,tx);
 687             fiy2             = _mm_add_pd(fiy2,ty);
 688             fiz2             = _mm_add_pd(fiz2,tz);
 689
 690             fjx0             = _mm_add_pd(fjx0,tx);
 691             fjy0             = _mm_add_pd(fjy0,ty);
 692             fjz0             = _mm_add_pd(fjz0,tz);
 693
 694             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 695
 696             /* Inner loop uses 84 flops */
 697         }
 698
 699         if(jidx<j_index_end)
 700         {
 701
 702             jnrA             = jjnr[jidx];
 703             j_coord_offsetA  = DIM*jnrA;
 704
 705             /* load j atom coordinates */
 706             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 707                                               &jx0,&jy0,&jz0);
 708
 709             /* Calculate displacement vector */
 710             dx00             = _mm_sub_pd(ix0,jx0);
 711             dy00             = _mm_sub_pd(iy0,jy0);
 712             dz00             = _mm_sub_pd(iz0,jz0);
 713             dx10             = _mm_sub_pd(ix1,jx0);
 714             dy10             = _mm_sub_pd(iy1,jy0);
 715             dz10             = _mm_sub_pd(iz1,jz0);
 716             dx20             = _mm_sub_pd(ix2,jx0);
 717             dy20             = _mm_sub_pd(iy2,jy0);
 718             dz20             = _mm_sub_pd(iz2,jz0);
 719
 720             /* Calculate squared distance and things based on it */
 721             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 722             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 723             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 724
 725             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 726             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 727             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 728
 729             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 730             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 731             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 732
 733             /* Load parameters for j particles */
 734             jq0              = _mm_load_sd(charge+jnrA+0);
 735
 736             fjx0             = _mm_setzero_pd();
 737             fjy0             = _mm_setzero_pd();
 738             fjz0             = _mm_setzero_pd();
 739
 740             /**************************
 741              * CALCULATE INTERACTIONS *
 742              **************************/
 743
 744             /* Compute parameters for interactions between i and j atoms */
 745             qq00             = _mm_mul_pd(iq0,jq0);
 746
 747             /* COULOMB ELECTROSTATICS */
 748             velec            = _mm_mul_pd(qq00,rinv00);
 749             felec            = _mm_mul_pd(velec,rinvsq00);
 750
 751             fscal            = felec;
 752
 753             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 754
 755             /* Calculate temporary vectorial force */
 756             tx               = _mm_mul_pd(fscal,dx00);
 757             ty               = _mm_mul_pd(fscal,dy00);
 758             tz               = _mm_mul_pd(fscal,dz00);
 759
 760             /* Update vectorial force */
 761             fix0             = _mm_add_pd(fix0,tx);
 762             fiy0             = _mm_add_pd(fiy0,ty);
 763             fiz0             = _mm_add_pd(fiz0,tz);
 764
 765             fjx0             = _mm_add_pd(fjx0,tx);
 766             fjy0             = _mm_add_pd(fjy0,ty);
 767             fjz0             = _mm_add_pd(fjz0,tz);
 768
 769             /**************************
 770              * CALCULATE INTERACTIONS *
 771              **************************/
 772
 773             /* Compute parameters for interactions between i and j atoms */
 774             qq10             = _mm_mul_pd(iq1,jq0);
 775
 776             /* COULOMB ELECTROSTATICS */
 777             velec            = _mm_mul_pd(qq10,rinv10);
 778             felec            = _mm_mul_pd(velec,rinvsq10);
 779
 780             fscal            = felec;
 781
 782             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 783
 784             /* Calculate temporary vectorial force */
 785             tx               = _mm_mul_pd(fscal,dx10);
 786             ty               = _mm_mul_pd(fscal,dy10);
 787             tz               = _mm_mul_pd(fscal,dz10);
 788
 789             /* Update vectorial force */
 790             fix1             = _mm_add_pd(fix1,tx);
 791             fiy1             = _mm_add_pd(fiy1,ty);
 792             fiz1             = _mm_add_pd(fiz1,tz);
 793
 794             fjx0             = _mm_add_pd(fjx0,tx);
 795             fjy0             = _mm_add_pd(fjy0,ty);
 796             fjz0             = _mm_add_pd(fjz0,tz);
 797
 798             /**************************
 799              * CALCULATE INTERACTIONS *
 800              **************************/
 801
 802             /* Compute parameters for interactions between i and j atoms */
 803             qq20             = _mm_mul_pd(iq2,jq0);
 804
 805             /* COULOMB ELECTROSTATICS */
 806             velec            = _mm_mul_pd(qq20,rinv20);
 807             felec            = _mm_mul_pd(velec,rinvsq20);
 808
 809             fscal            = felec;
 810
 811             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 812
 813             /* Calculate temporary vectorial force */
 814             tx               = _mm_mul_pd(fscal,dx20);
 815             ty               = _mm_mul_pd(fscal,dy20);
 816             tz               = _mm_mul_pd(fscal,dz20);
 817
 818             /* Update vectorial force */
 819             fix2             = _mm_add_pd(fix2,tx);
 820             fiy2             = _mm_add_pd(fiy2,ty);
 821             fiz2             = _mm_add_pd(fiz2,tz);
 822
 823             fjx0             = _mm_add_pd(fjx0,tx);
 824             fjy0             = _mm_add_pd(fjy0,ty);
 825             fjz0             = _mm_add_pd(fjz0,tz);
 826
 827             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 828
 829             /* Inner loop uses 84 flops */
 830         }
 831
 832         /* End of innermost loop */
 833
 834         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 835                                               f+i_coord_offset,fshift+i_shift_offset);
 836
 837         /* Increment number of inner iterations */
 838         inneriter                  += j_index_end - j_index_start;
 839
 840         /* Outer loop uses 18 flops */
 841     }
 842
 843     /* Increment number of outer iterations */
 844     outeriter        += nri;
 845
 846     /* Update outer/inner flops */
 847
 848     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*84);
 849 }