src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecRF_VdwNone_GeomW3P1_sse4_1_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse4_1_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomW3P1_VF_sse4_1_double
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            None
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRF_VdwNone_GeomW3P1_VF_sse4_1_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwioffset1;
  82     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  83     int              vdwioffset2;
  84     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  85     int              vdwjidx0A,vdwjidx0B;
  86     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  89     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  90     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     __m128d          dummy_mask,cutoff_mask;
  93     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  94     __m128d          one     = _mm_set1_pd(1.0);
  95     __m128d          two     = _mm_set1_pd(2.0);
  96     x                = xx[0];
  97     f                = ff[0];
  98
  99     nri              = nlist->nri;
 100     iinr             = nlist->iinr;
 101     jindex           = nlist->jindex;
 102     jjnr             = nlist->jjnr;
 103     shiftidx         = nlist->shift;
 104     gid              = nlist->gid;
 105     shiftvec         = fr->shift_vec[0];
 106     fshift           = fr->fshift[0];
 107     facel            = _mm_set1_pd(fr->ic->epsfac);
 108     charge           = mdatoms->chargeA;
 109     krf              = _mm_set1_pd(fr->ic->k_rf);
 110     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 111     crf              = _mm_set1_pd(fr->ic->c_rf);
 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           = sse41_invsqrt_d(rsq00);
 189             rinv10           = sse41_invsqrt_d(rsq10);
 190             rinv20           = sse41_invsqrt_d(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             /* REACTION-FIELD ELECTROSTATICS */
 211             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 212             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 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             /* REACTION-FIELD ELECTROSTATICS */
 241             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 242             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 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             /* REACTION-FIELD ELECTROSTATICS */
 271             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 272             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 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 99 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           = sse41_invsqrt_d(rsq00);
 325             rinv10           = sse41_invsqrt_d(rsq10);
 326             rinv20           = sse41_invsqrt_d(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             /* REACTION-FIELD ELECTROSTATICS */
 347             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 348             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 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             /* REACTION-FIELD ELECTROSTATICS */
 380             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 381             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 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             /* REACTION-FIELD ELECTROSTATICS */
 413             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 414             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 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 99 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*99);
 464 }
 465 /*
 466  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomW3P1_F_sse4_1_double
 467  * Electrostatics interaction: ReactionField
 468  * VdW interaction:            None
 469  * Geometry:                   Water3-Particle
 470  * Calculate force/pot:        Force
 471  */
 472 void
 473 nb_kernel_ElecRF_VdwNone_GeomW3P1_F_sse4_1_double
 474                     (t_nblist                    * gmx_restrict       nlist,
 475                      rvec                        * gmx_restrict          xx,
 476                      rvec                        * gmx_restrict          ff,
 477                      struct 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->ic->epsfac);
 524     charge           = mdatoms->chargeA;
 525     krf              = _mm_set1_pd(fr->ic->k_rf);
 526     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 527     crf              = _mm_set1_pd(fr->ic->c_rf);
 528
 529     /* Setup water-specific parameters */
 530     inr              = nlist->iinr[0];
 531     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 532     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 533     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 534
 535     /* Avoid stupid compiler warnings */
 536     jnrA = jnrB = 0;
 537     j_coord_offsetA = 0;
 538     j_coord_offsetB = 0;
 539
 540     outeriter        = 0;
 541     inneriter        = 0;
 542
 543     /* Start outer loop over neighborlists */
 544     for(iidx=0; iidx<nri; iidx++)
 545     {
 546         /* Load shift vector for this list */
 547         i_shift_offset   = DIM*shiftidx[iidx];
 548
 549         /* Load limits for loop over neighbors */
 550         j_index_start    = jindex[iidx];
 551         j_index_end      = jindex[iidx+1];
 552
 553         /* Get outer coordinate index */
 554         inr              = iinr[iidx];
 555         i_coord_offset   = DIM*inr;
 556
 557         /* Load i particle coords and add shift vector */
 558         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 559                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 560
 561         fix0             = _mm_setzero_pd();
 562         fiy0             = _mm_setzero_pd();
 563         fiz0             = _mm_setzero_pd();
 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
 571         /* Start inner kernel loop */
 572         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 573         {
 574
 575             /* Get j neighbor index, and coordinate index */
 576             jnrA             = jjnr[jidx];
 577             jnrB             = jjnr[jidx+1];
 578             j_coord_offsetA  = DIM*jnrA;
 579             j_coord_offsetB  = DIM*jnrB;
 580
 581             /* load j atom coordinates */
 582             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 583                                               &jx0,&jy0,&jz0);
 584
 585             /* Calculate displacement vector */
 586             dx00             = _mm_sub_pd(ix0,jx0);
 587             dy00             = _mm_sub_pd(iy0,jy0);
 588             dz00             = _mm_sub_pd(iz0,jz0);
 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
 596             /* Calculate squared distance and things based on it */
 597             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 598             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 599             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 600
 601             rinv00           = sse41_invsqrt_d(rsq00);
 602             rinv10           = sse41_invsqrt_d(rsq10);
 603             rinv20           = sse41_invsqrt_d(rsq20);
 604
 605             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 606             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 607             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 608
 609             /* Load parameters for j particles */
 610             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 611
 612             fjx0             = _mm_setzero_pd();
 613             fjy0             = _mm_setzero_pd();
 614             fjz0             = _mm_setzero_pd();
 615
 616             /**************************
 617              * CALCULATE INTERACTIONS *
 618              **************************/
 619
 620             /* Compute parameters for interactions between i and j atoms */
 621             qq00             = _mm_mul_pd(iq0,jq0);
 622
 623             /* REACTION-FIELD ELECTROSTATICS */
 624             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 625
 626             fscal            = felec;
 627
 628             /* Calculate temporary vectorial force */
 629             tx               = _mm_mul_pd(fscal,dx00);
 630             ty               = _mm_mul_pd(fscal,dy00);
 631             tz               = _mm_mul_pd(fscal,dz00);
 632
 633             /* Update vectorial force */
 634             fix0             = _mm_add_pd(fix0,tx);
 635             fiy0             = _mm_add_pd(fiy0,ty);
 636             fiz0             = _mm_add_pd(fiz0,tz);
 637
 638             fjx0             = _mm_add_pd(fjx0,tx);
 639             fjy0             = _mm_add_pd(fjy0,ty);
 640             fjz0             = _mm_add_pd(fjz0,tz);
 641
 642             /**************************
 643              * CALCULATE INTERACTIONS *
 644              **************************/
 645
 646             /* Compute parameters for interactions between i and j atoms */
 647             qq10             = _mm_mul_pd(iq1,jq0);
 648
 649             /* REACTION-FIELD ELECTROSTATICS */
 650             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 651
 652             fscal            = felec;
 653
 654             /* Calculate temporary vectorial force */
 655             tx               = _mm_mul_pd(fscal,dx10);
 656             ty               = _mm_mul_pd(fscal,dy10);
 657             tz               = _mm_mul_pd(fscal,dz10);
 658
 659             /* Update vectorial force */
 660             fix1             = _mm_add_pd(fix1,tx);
 661             fiy1             = _mm_add_pd(fiy1,ty);
 662             fiz1             = _mm_add_pd(fiz1,tz);
 663
 664             fjx0             = _mm_add_pd(fjx0,tx);
 665             fjy0             = _mm_add_pd(fjy0,ty);
 666             fjz0             = _mm_add_pd(fjz0,tz);
 667
 668             /**************************
 669              * CALCULATE INTERACTIONS *
 670              **************************/
 671
 672             /* Compute parameters for interactions between i and j atoms */
 673             qq20             = _mm_mul_pd(iq2,jq0);
 674
 675             /* REACTION-FIELD ELECTROSTATICS */
 676             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 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           = sse41_invsqrt_d(rsq00);
 726             rinv10           = sse41_invsqrt_d(rsq10);
 727             rinv20           = sse41_invsqrt_d(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             /* REACTION-FIELD ELECTROSTATICS */
 748             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 749
 750             fscal            = felec;
 751
 752             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 753
 754             /* Calculate temporary vectorial force */
 755             tx               = _mm_mul_pd(fscal,dx00);
 756             ty               = _mm_mul_pd(fscal,dy00);
 757             tz               = _mm_mul_pd(fscal,dz00);
 758
 759             /* Update vectorial force */
 760             fix0             = _mm_add_pd(fix0,tx);
 761             fiy0             = _mm_add_pd(fiy0,ty);
 762             fiz0             = _mm_add_pd(fiz0,tz);
 763
 764             fjx0             = _mm_add_pd(fjx0,tx);
 765             fjy0             = _mm_add_pd(fjy0,ty);
 766             fjz0             = _mm_add_pd(fjz0,tz);
 767
 768             /**************************
 769              * CALCULATE INTERACTIONS *
 770              **************************/
 771
 772             /* Compute parameters for interactions between i and j atoms */
 773             qq10             = _mm_mul_pd(iq1,jq0);
 774
 775             /* REACTION-FIELD ELECTROSTATICS */
 776             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 777
 778             fscal            = felec;
 779
 780             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 781
 782             /* Calculate temporary vectorial force */
 783             tx               = _mm_mul_pd(fscal,dx10);
 784             ty               = _mm_mul_pd(fscal,dy10);
 785             tz               = _mm_mul_pd(fscal,dz10);
 786
 787             /* Update vectorial force */
 788             fix1             = _mm_add_pd(fix1,tx);
 789             fiy1             = _mm_add_pd(fiy1,ty);
 790             fiz1             = _mm_add_pd(fiz1,tz);
 791
 792             fjx0             = _mm_add_pd(fjx0,tx);
 793             fjy0             = _mm_add_pd(fjy0,ty);
 794             fjz0             = _mm_add_pd(fjz0,tz);
 795
 796             /**************************
 797              * CALCULATE INTERACTIONS *
 798              **************************/
 799
 800             /* Compute parameters for interactions between i and j atoms */
 801             qq20             = _mm_mul_pd(iq2,jq0);
 802
 803             /* REACTION-FIELD ELECTROSTATICS */
 804             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 805
 806             fscal            = felec;
 807
 808             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 809
 810             /* Calculate temporary vectorial force */
 811             tx               = _mm_mul_pd(fscal,dx20);
 812             ty               = _mm_mul_pd(fscal,dy20);
 813             tz               = _mm_mul_pd(fscal,dz20);
 814
 815             /* Update vectorial force */
 816             fix2             = _mm_add_pd(fix2,tx);
 817             fiy2             = _mm_add_pd(fiy2,ty);
 818             fiz2             = _mm_add_pd(fiz2,tz);
 819
 820             fjx0             = _mm_add_pd(fjx0,tx);
 821             fjy0             = _mm_add_pd(fjy0,ty);
 822             fjz0             = _mm_add_pd(fjz0,tz);
 823
 824             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 825
 826             /* Inner loop uses 84 flops */
 827         }
 828
 829         /* End of innermost loop */
 830
 831         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 832                                               f+i_coord_offset,fshift+i_shift_offset);
 833
 834         /* Increment number of inner iterations */
 835         inneriter                  += j_index_end - j_index_start;
 836
 837         /* Outer loop uses 18 flops */
 838     }
 839
 840     /* Increment number of outer iterations */
 841     outeriter        += nri;
 842
 843     /* Update outer/inner flops */
 844
 845     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*84);
 846 }