src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecRF_VdwNone_GeomW4P1_sse4_1_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gmx_math_x86_sse4_1_double.h"
  50 #include "kernelutil_x86_sse4_1_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomW4P1_VF_sse4_1_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            None
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRF_VdwNone_GeomW4P1_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              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          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  91     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  92     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  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     krf              = _mm_set1_pd(fr->ic->k_rf);
 113     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 114     crf              = _mm_set1_pd(fr->ic->c_rf);
 115
 116     /* Setup water-specific parameters */
 117     inr              = nlist->iinr[0];
 118     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 119     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 120     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 121
 122     /* Avoid stupid compiler warnings */
 123     jnrA = jnrB = 0;
 124     j_coord_offsetA = 0;
 125     j_coord_offsetB = 0;
 126
 127     outeriter        = 0;
 128     inneriter        = 0;
 129
 130     /* Start outer loop over neighborlists */
 131     for(iidx=0; iidx<nri; iidx++)
 132     {
 133         /* Load shift vector for this list */
 134         i_shift_offset   = DIM*shiftidx[iidx];
 135
 136         /* Load limits for loop over neighbors */
 137         j_index_start    = jindex[iidx];
 138         j_index_end      = jindex[iidx+1];
 139
 140         /* Get outer coordinate index */
 141         inr              = iinr[iidx];
 142         i_coord_offset   = DIM*inr;
 143
 144         /* Load i particle coords and add shift vector */
 145         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 146                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 147
 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         fix3             = _mm_setzero_pd();
 155         fiy3             = _mm_setzero_pd();
 156         fiz3             = _mm_setzero_pd();
 157
 158         /* Reset potential sums */
 159         velecsum         = _mm_setzero_pd();
 160
 161         /* Start inner kernel loop */
 162         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 163         {
 164
 165             /* Get j neighbor index, and coordinate index */
 166             jnrA             = jjnr[jidx];
 167             jnrB             = jjnr[jidx+1];
 168             j_coord_offsetA  = DIM*jnrA;
 169             j_coord_offsetB  = DIM*jnrB;
 170
 171             /* load j atom coordinates */
 172             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 173                                               &jx0,&jy0,&jz0);
 174
 175             /* Calculate displacement vector */
 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             dx30             = _mm_sub_pd(ix3,jx0);
 183             dy30             = _mm_sub_pd(iy3,jy0);
 184             dz30             = _mm_sub_pd(iz3,jz0);
 185
 186             /* Calculate squared distance and things based on it */
 187             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 188             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 189             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 190
 191             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 192             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 193             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 194
 195             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 196             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 197             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 198
 199             /* Load parameters for j particles */
 200             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 201
 202             fjx0             = _mm_setzero_pd();
 203             fjy0             = _mm_setzero_pd();
 204             fjz0             = _mm_setzero_pd();
 205
 206             /**************************
 207              * CALCULATE INTERACTIONS *
 208              **************************/
 209
 210             /* Compute parameters for interactions between i and j atoms */
 211             qq10             = _mm_mul_pd(iq1,jq0);
 212
 213             /* REACTION-FIELD ELECTROSTATICS */
 214             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 215             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 216
 217             /* Update potential sum for this i atom from the interaction with this j atom. */
 218             velecsum         = _mm_add_pd(velecsum,velec);
 219
 220             fscal            = felec;
 221
 222             /* Calculate temporary vectorial force */
 223             tx               = _mm_mul_pd(fscal,dx10);
 224             ty               = _mm_mul_pd(fscal,dy10);
 225             tz               = _mm_mul_pd(fscal,dz10);
 226
 227             /* Update vectorial force */
 228             fix1             = _mm_add_pd(fix1,tx);
 229             fiy1             = _mm_add_pd(fiy1,ty);
 230             fiz1             = _mm_add_pd(fiz1,tz);
 231
 232             fjx0             = _mm_add_pd(fjx0,tx);
 233             fjy0             = _mm_add_pd(fjy0,ty);
 234             fjz0             = _mm_add_pd(fjz0,tz);
 235
 236             /**************************
 237              * CALCULATE INTERACTIONS *
 238              **************************/
 239
 240             /* Compute parameters for interactions between i and j atoms */
 241             qq20             = _mm_mul_pd(iq2,jq0);
 242
 243             /* REACTION-FIELD ELECTROSTATICS */
 244             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 245             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 246
 247             /* Update potential sum for this i atom from the interaction with this j atom. */
 248             velecsum         = _mm_add_pd(velecsum,velec);
 249
 250             fscal            = felec;
 251
 252             /* Calculate temporary vectorial force */
 253             tx               = _mm_mul_pd(fscal,dx20);
 254             ty               = _mm_mul_pd(fscal,dy20);
 255             tz               = _mm_mul_pd(fscal,dz20);
 256
 257             /* Update vectorial force */
 258             fix2             = _mm_add_pd(fix2,tx);
 259             fiy2             = _mm_add_pd(fiy2,ty);
 260             fiz2             = _mm_add_pd(fiz2,tz);
 261
 262             fjx0             = _mm_add_pd(fjx0,tx);
 263             fjy0             = _mm_add_pd(fjy0,ty);
 264             fjz0             = _mm_add_pd(fjz0,tz);
 265
 266             /**************************
 267              * CALCULATE INTERACTIONS *
 268              **************************/
 269
 270             /* Compute parameters for interactions between i and j atoms */
 271             qq30             = _mm_mul_pd(iq3,jq0);
 272
 273             /* REACTION-FIELD ELECTROSTATICS */
 274             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_add_pd(rinv30,_mm_mul_pd(krf,rsq30)),crf));
 275             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
 276
 277             /* Update potential sum for this i atom from the interaction with this j atom. */
 278             velecsum         = _mm_add_pd(velecsum,velec);
 279
 280             fscal            = felec;
 281
 282             /* Calculate temporary vectorial force */
 283             tx               = _mm_mul_pd(fscal,dx30);
 284             ty               = _mm_mul_pd(fscal,dy30);
 285             tz               = _mm_mul_pd(fscal,dz30);
 286
 287             /* Update vectorial force */
 288             fix3             = _mm_add_pd(fix3,tx);
 289             fiy3             = _mm_add_pd(fiy3,ty);
 290             fiz3             = _mm_add_pd(fiz3,tz);
 291
 292             fjx0             = _mm_add_pd(fjx0,tx);
 293             fjy0             = _mm_add_pd(fjy0,ty);
 294             fjz0             = _mm_add_pd(fjz0,tz);
 295
 296             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 297
 298             /* Inner loop uses 99 flops */
 299         }
 300
 301         if(jidx<j_index_end)
 302         {
 303
 304             jnrA             = jjnr[jidx];
 305             j_coord_offsetA  = DIM*jnrA;
 306
 307             /* load j atom coordinates */
 308             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 309                                               &jx0,&jy0,&jz0);
 310
 311             /* Calculate displacement vector */
 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             dx30             = _mm_sub_pd(ix3,jx0);
 319             dy30             = _mm_sub_pd(iy3,jy0);
 320             dz30             = _mm_sub_pd(iz3,jz0);
 321
 322             /* Calculate squared distance and things based on it */
 323             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 324             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 325             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 326
 327             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 328             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 329             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 330
 331             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 332             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 333             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 334
 335             /* Load parameters for j particles */
 336             jq0              = _mm_load_sd(charge+jnrA+0);
 337
 338             fjx0             = _mm_setzero_pd();
 339             fjy0             = _mm_setzero_pd();
 340             fjz0             = _mm_setzero_pd();
 341
 342             /**************************
 343              * CALCULATE INTERACTIONS *
 344              **************************/
 345
 346             /* Compute parameters for interactions between i and j atoms */
 347             qq10             = _mm_mul_pd(iq1,jq0);
 348
 349             /* REACTION-FIELD ELECTROSTATICS */
 350             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 351             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 352
 353             /* Update potential sum for this i atom from the interaction with this j atom. */
 354             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 355             velecsum         = _mm_add_pd(velecsum,velec);
 356
 357             fscal            = felec;
 358
 359             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 360
 361             /* Calculate temporary vectorial force */
 362             tx               = _mm_mul_pd(fscal,dx10);
 363             ty               = _mm_mul_pd(fscal,dy10);
 364             tz               = _mm_mul_pd(fscal,dz10);
 365
 366             /* Update vectorial force */
 367             fix1             = _mm_add_pd(fix1,tx);
 368             fiy1             = _mm_add_pd(fiy1,ty);
 369             fiz1             = _mm_add_pd(fiz1,tz);
 370
 371             fjx0             = _mm_add_pd(fjx0,tx);
 372             fjy0             = _mm_add_pd(fjy0,ty);
 373             fjz0             = _mm_add_pd(fjz0,tz);
 374
 375             /**************************
 376              * CALCULATE INTERACTIONS *
 377              **************************/
 378
 379             /* Compute parameters for interactions between i and j atoms */
 380             qq20             = _mm_mul_pd(iq2,jq0);
 381
 382             /* REACTION-FIELD ELECTROSTATICS */
 383             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 384             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 385
 386             /* Update potential sum for this i atom from the interaction with this j atom. */
 387             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 388             velecsum         = _mm_add_pd(velecsum,velec);
 389
 390             fscal            = felec;
 391
 392             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 393
 394             /* Calculate temporary vectorial force */
 395             tx               = _mm_mul_pd(fscal,dx20);
 396             ty               = _mm_mul_pd(fscal,dy20);
 397             tz               = _mm_mul_pd(fscal,dz20);
 398
 399             /* Update vectorial force */
 400             fix2             = _mm_add_pd(fix2,tx);
 401             fiy2             = _mm_add_pd(fiy2,ty);
 402             fiz2             = _mm_add_pd(fiz2,tz);
 403
 404             fjx0             = _mm_add_pd(fjx0,tx);
 405             fjy0             = _mm_add_pd(fjy0,ty);
 406             fjz0             = _mm_add_pd(fjz0,tz);
 407
 408             /**************************
 409              * CALCULATE INTERACTIONS *
 410              **************************/
 411
 412             /* Compute parameters for interactions between i and j atoms */
 413             qq30             = _mm_mul_pd(iq3,jq0);
 414
 415             /* REACTION-FIELD ELECTROSTATICS */
 416             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_add_pd(rinv30,_mm_mul_pd(krf,rsq30)),crf));
 417             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
 418
 419             /* Update potential sum for this i atom from the interaction with this j atom. */
 420             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 421             velecsum         = _mm_add_pd(velecsum,velec);
 422
 423             fscal            = felec;
 424
 425             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 426
 427             /* Calculate temporary vectorial force */
 428             tx               = _mm_mul_pd(fscal,dx30);
 429             ty               = _mm_mul_pd(fscal,dy30);
 430             tz               = _mm_mul_pd(fscal,dz30);
 431
 432             /* Update vectorial force */
 433             fix3             = _mm_add_pd(fix3,tx);
 434             fiy3             = _mm_add_pd(fiy3,ty);
 435             fiz3             = _mm_add_pd(fiz3,tz);
 436
 437             fjx0             = _mm_add_pd(fjx0,tx);
 438             fjy0             = _mm_add_pd(fjy0,ty);
 439             fjz0             = _mm_add_pd(fjz0,tz);
 440
 441             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 442
 443             /* Inner loop uses 99 flops */
 444         }
 445
 446         /* End of innermost loop */
 447
 448         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 449                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 450
 451         ggid                        = gid[iidx];
 452         /* Update potential energies */
 453         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 454
 455         /* Increment number of inner iterations */
 456         inneriter                  += j_index_end - j_index_start;
 457
 458         /* Outer loop uses 19 flops */
 459     }
 460
 461     /* Increment number of outer iterations */
 462     outeriter        += nri;
 463
 464     /* Update outer/inner flops */
 465
 466     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*99);
 467 }
 468 /*
 469  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomW4P1_F_sse4_1_double
 470  * Electrostatics interaction: ReactionField
 471  * VdW interaction:            None
 472  * Geometry:                   Water4-Particle
 473  * Calculate force/pot:        Force
 474  */
 475 void
 476 nb_kernel_ElecRF_VdwNone_GeomW4P1_F_sse4_1_double
 477                     (t_nblist                    * gmx_restrict       nlist,
 478                      rvec                        * gmx_restrict          xx,
 479                      rvec                        * gmx_restrict          ff,
 480                      t_forcerec                  * gmx_restrict          fr,
 481                      t_mdatoms                   * gmx_restrict     mdatoms,
 482                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 483                      t_nrnb                      * gmx_restrict        nrnb)
 484 {
 485     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 486      * just 0 for non-waters.
 487      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 488      * jnr indices corresponding to data put in the four positions in the SIMD register.
 489      */
 490     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 491     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 492     int              jnrA,jnrB;
 493     int              j_coord_offsetA,j_coord_offsetB;
 494     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 495     real             rcutoff_scalar;
 496     real             *shiftvec,*fshift,*x,*f;
 497     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 498     int              vdwioffset1;
 499     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 500     int              vdwioffset2;
 501     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 502     int              vdwioffset3;
 503     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 504     int              vdwjidx0A,vdwjidx0B;
 505     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 506     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 507     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 508     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 509     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 510     real             *charge;
 511     __m128d          dummy_mask,cutoff_mask;
 512     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 513     __m128d          one     = _mm_set1_pd(1.0);
 514     __m128d          two     = _mm_set1_pd(2.0);
 515     x                = xx[0];
 516     f                = ff[0];
 517
 518     nri              = nlist->nri;
 519     iinr             = nlist->iinr;
 520     jindex           = nlist->jindex;
 521     jjnr             = nlist->jjnr;
 522     shiftidx         = nlist->shift;
 523     gid              = nlist->gid;
 524     shiftvec         = fr->shift_vec[0];
 525     fshift           = fr->fshift[0];
 526     facel            = _mm_set1_pd(fr->epsfac);
 527     charge           = mdatoms->chargeA;
 528     krf              = _mm_set1_pd(fr->ic->k_rf);
 529     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 530     crf              = _mm_set1_pd(fr->ic->c_rf);
 531
 532     /* Setup water-specific parameters */
 533     inr              = nlist->iinr[0];
 534     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 535     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 536     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 537
 538     /* Avoid stupid compiler warnings */
 539     jnrA = jnrB = 0;
 540     j_coord_offsetA = 0;
 541     j_coord_offsetB = 0;
 542
 543     outeriter        = 0;
 544     inneriter        = 0;
 545
 546     /* Start outer loop over neighborlists */
 547     for(iidx=0; iidx<nri; iidx++)
 548     {
 549         /* Load shift vector for this list */
 550         i_shift_offset   = DIM*shiftidx[iidx];
 551
 552         /* Load limits for loop over neighbors */
 553         j_index_start    = jindex[iidx];
 554         j_index_end      = jindex[iidx+1];
 555
 556         /* Get outer coordinate index */
 557         inr              = iinr[iidx];
 558         i_coord_offset   = DIM*inr;
 559
 560         /* Load i particle coords and add shift vector */
 561         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 562                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 563
 564         fix1             = _mm_setzero_pd();
 565         fiy1             = _mm_setzero_pd();
 566         fiz1             = _mm_setzero_pd();
 567         fix2             = _mm_setzero_pd();
 568         fiy2             = _mm_setzero_pd();
 569         fiz2             = _mm_setzero_pd();
 570         fix3             = _mm_setzero_pd();
 571         fiy3             = _mm_setzero_pd();
 572         fiz3             = _mm_setzero_pd();
 573
 574         /* Start inner kernel loop */
 575         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 576         {
 577
 578             /* Get j neighbor index, and coordinate index */
 579             jnrA             = jjnr[jidx];
 580             jnrB             = jjnr[jidx+1];
 581             j_coord_offsetA  = DIM*jnrA;
 582             j_coord_offsetB  = DIM*jnrB;
 583
 584             /* load j atom coordinates */
 585             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 586                                               &jx0,&jy0,&jz0);
 587
 588             /* Calculate displacement vector */
 589             dx10             = _mm_sub_pd(ix1,jx0);
 590             dy10             = _mm_sub_pd(iy1,jy0);
 591             dz10             = _mm_sub_pd(iz1,jz0);
 592             dx20             = _mm_sub_pd(ix2,jx0);
 593             dy20             = _mm_sub_pd(iy2,jy0);
 594             dz20             = _mm_sub_pd(iz2,jz0);
 595             dx30             = _mm_sub_pd(ix3,jx0);
 596             dy30             = _mm_sub_pd(iy3,jy0);
 597             dz30             = _mm_sub_pd(iz3,jz0);
 598
 599             /* Calculate squared distance and things based on it */
 600             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 601             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 602             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 603
 604             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 605             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 606             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 607
 608             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 609             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 610             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 611
 612             /* Load parameters for j particles */
 613             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 614
 615             fjx0             = _mm_setzero_pd();
 616             fjy0             = _mm_setzero_pd();
 617             fjz0             = _mm_setzero_pd();
 618
 619             /**************************
 620              * CALCULATE INTERACTIONS *
 621              **************************/
 622
 623             /* Compute parameters for interactions between i and j atoms */
 624             qq10             = _mm_mul_pd(iq1,jq0);
 625
 626             /* REACTION-FIELD ELECTROSTATICS */
 627             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 628
 629             fscal            = felec;
 630
 631             /* Calculate temporary vectorial force */
 632             tx               = _mm_mul_pd(fscal,dx10);
 633             ty               = _mm_mul_pd(fscal,dy10);
 634             tz               = _mm_mul_pd(fscal,dz10);
 635
 636             /* Update vectorial force */
 637             fix1             = _mm_add_pd(fix1,tx);
 638             fiy1             = _mm_add_pd(fiy1,ty);
 639             fiz1             = _mm_add_pd(fiz1,tz);
 640
 641             fjx0             = _mm_add_pd(fjx0,tx);
 642             fjy0             = _mm_add_pd(fjy0,ty);
 643             fjz0             = _mm_add_pd(fjz0,tz);
 644
 645             /**************************
 646              * CALCULATE INTERACTIONS *
 647              **************************/
 648
 649             /* Compute parameters for interactions between i and j atoms */
 650             qq20             = _mm_mul_pd(iq2,jq0);
 651
 652             /* REACTION-FIELD ELECTROSTATICS */
 653             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 654
 655             fscal            = felec;
 656
 657             /* Calculate temporary vectorial force */
 658             tx               = _mm_mul_pd(fscal,dx20);
 659             ty               = _mm_mul_pd(fscal,dy20);
 660             tz               = _mm_mul_pd(fscal,dz20);
 661
 662             /* Update vectorial force */
 663             fix2             = _mm_add_pd(fix2,tx);
 664             fiy2             = _mm_add_pd(fiy2,ty);
 665             fiz2             = _mm_add_pd(fiz2,tz);
 666
 667             fjx0             = _mm_add_pd(fjx0,tx);
 668             fjy0             = _mm_add_pd(fjy0,ty);
 669             fjz0             = _mm_add_pd(fjz0,tz);
 670
 671             /**************************
 672              * CALCULATE INTERACTIONS *
 673              **************************/
 674
 675             /* Compute parameters for interactions between i and j atoms */
 676             qq30             = _mm_mul_pd(iq3,jq0);
 677
 678             /* REACTION-FIELD ELECTROSTATICS */
 679             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
 680
 681             fscal            = felec;
 682
 683             /* Calculate temporary vectorial force */
 684             tx               = _mm_mul_pd(fscal,dx30);
 685             ty               = _mm_mul_pd(fscal,dy30);
 686             tz               = _mm_mul_pd(fscal,dz30);
 687
 688             /* Update vectorial force */
 689             fix3             = _mm_add_pd(fix3,tx);
 690             fiy3             = _mm_add_pd(fiy3,ty);
 691             fiz3             = _mm_add_pd(fiz3,tz);
 692
 693             fjx0             = _mm_add_pd(fjx0,tx);
 694             fjy0             = _mm_add_pd(fjy0,ty);
 695             fjz0             = _mm_add_pd(fjz0,tz);
 696
 697             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 698
 699             /* Inner loop uses 84 flops */
 700         }
 701
 702         if(jidx<j_index_end)
 703         {
 704
 705             jnrA             = jjnr[jidx];
 706             j_coord_offsetA  = DIM*jnrA;
 707
 708             /* load j atom coordinates */
 709             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 710                                               &jx0,&jy0,&jz0);
 711
 712             /* Calculate displacement vector */
 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             dx30             = _mm_sub_pd(ix3,jx0);
 720             dy30             = _mm_sub_pd(iy3,jy0);
 721             dz30             = _mm_sub_pd(iz3,jz0);
 722
 723             /* Calculate squared distance and things based on it */
 724             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 725             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 726             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 727
 728             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 729             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 730             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 731
 732             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 733             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 734             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 735
 736             /* Load parameters for j particles */
 737             jq0              = _mm_load_sd(charge+jnrA+0);
 738
 739             fjx0             = _mm_setzero_pd();
 740             fjy0             = _mm_setzero_pd();
 741             fjz0             = _mm_setzero_pd();
 742
 743             /**************************
 744              * CALCULATE INTERACTIONS *
 745              **************************/
 746
 747             /* Compute parameters for interactions between i and j atoms */
 748             qq10             = _mm_mul_pd(iq1,jq0);
 749
 750             /* REACTION-FIELD ELECTROSTATICS */
 751             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 752
 753             fscal            = felec;
 754
 755             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 756
 757             /* Calculate temporary vectorial force */
 758             tx               = _mm_mul_pd(fscal,dx10);
 759             ty               = _mm_mul_pd(fscal,dy10);
 760             tz               = _mm_mul_pd(fscal,dz10);
 761
 762             /* Update vectorial force */
 763             fix1             = _mm_add_pd(fix1,tx);
 764             fiy1             = _mm_add_pd(fiy1,ty);
 765             fiz1             = _mm_add_pd(fiz1,tz);
 766
 767             fjx0             = _mm_add_pd(fjx0,tx);
 768             fjy0             = _mm_add_pd(fjy0,ty);
 769             fjz0             = _mm_add_pd(fjz0,tz);
 770
 771             /**************************
 772              * CALCULATE INTERACTIONS *
 773              **************************/
 774
 775             /* Compute parameters for interactions between i and j atoms */
 776             qq20             = _mm_mul_pd(iq2,jq0);
 777
 778             /* REACTION-FIELD ELECTROSTATICS */
 779             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 780
 781             fscal            = felec;
 782
 783             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 784
 785             /* Calculate temporary vectorial force */
 786             tx               = _mm_mul_pd(fscal,dx20);
 787             ty               = _mm_mul_pd(fscal,dy20);
 788             tz               = _mm_mul_pd(fscal,dz20);
 789
 790             /* Update vectorial force */
 791             fix2             = _mm_add_pd(fix2,tx);
 792             fiy2             = _mm_add_pd(fiy2,ty);
 793             fiz2             = _mm_add_pd(fiz2,tz);
 794
 795             fjx0             = _mm_add_pd(fjx0,tx);
 796             fjy0             = _mm_add_pd(fjy0,ty);
 797             fjz0             = _mm_add_pd(fjz0,tz);
 798
 799             /**************************
 800              * CALCULATE INTERACTIONS *
 801              **************************/
 802
 803             /* Compute parameters for interactions between i and j atoms */
 804             qq30             = _mm_mul_pd(iq3,jq0);
 805
 806             /* REACTION-FIELD ELECTROSTATICS */
 807             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
 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,dx30);
 815             ty               = _mm_mul_pd(fscal,dy30);
 816             tz               = _mm_mul_pd(fscal,dz30);
 817
 818             /* Update vectorial force */
 819             fix3             = _mm_add_pd(fix3,tx);
 820             fiy3             = _mm_add_pd(fiy3,ty);
 821             fiz3             = _mm_add_pd(fiz3,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(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 835                                               f+i_coord_offset+DIM,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_W4_F,outeriter*18 + inneriter*84);
 849 }