src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecRF_VdwNone_GeomW3P1_sse2_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_double.h"
  48 #include "kernelutil_x86_sse2_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomW3P1_VF_sse2_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            None
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRF_VdwNone_GeomW3P1_VF_sse2_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwjidx0A,vdwjidx0B;
  87     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  89     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  90     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  91     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     __m128d          dummy_mask,cutoff_mask;
  94     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  95     __m128d          one     = _mm_set1_pd(1.0);
  96     __m128d          two     = _mm_set1_pd(2.0);
  97     x                = xx[0];
  98     f                = ff[0];
  99
 100     nri              = nlist->nri;
 101     iinr             = nlist->iinr;
 102     jindex           = nlist->jindex;
 103     jjnr             = nlist->jjnr;
 104     shiftidx         = nlist->shift;
 105     gid              = nlist->gid;
 106     shiftvec         = fr->shift_vec[0];
 107     fshift           = fr->fshift[0];
 108     facel            = _mm_set1_pd(fr->epsfac);
 109     charge           = mdatoms->chargeA;
 110     krf              = _mm_set1_pd(fr->ic->k_rf);
 111     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 112     crf              = _mm_set1_pd(fr->ic->c_rf);
 113
 114     /* Setup water-specific parameters */
 115     inr              = nlist->iinr[0];
 116     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 117     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 118     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 119
 120     /* Avoid stupid compiler warnings */
 121     jnrA = jnrB = 0;
 122     j_coord_offsetA = 0;
 123     j_coord_offsetB = 0;
 124
 125     outeriter        = 0;
 126     inneriter        = 0;
 127
 128     /* Start outer loop over neighborlists */
 129     for(iidx=0; iidx<nri; iidx++)
 130     {
 131         /* Load shift vector for this list */
 132         i_shift_offset   = DIM*shiftidx[iidx];
 133
 134         /* Load limits for loop over neighbors */
 135         j_index_start    = jindex[iidx];
 136         j_index_end      = jindex[iidx+1];
 137
 138         /* Get outer coordinate index */
 139         inr              = iinr[iidx];
 140         i_coord_offset   = DIM*inr;
 141
 142         /* Load i particle coords and add shift vector */
 143         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 144                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 145
 146         fix0             = _mm_setzero_pd();
 147         fiy0             = _mm_setzero_pd();
 148         fiz0             = _mm_setzero_pd();
 149         fix1             = _mm_setzero_pd();
 150         fiy1             = _mm_setzero_pd();
 151         fiz1             = _mm_setzero_pd();
 152         fix2             = _mm_setzero_pd();
 153         fiy2             = _mm_setzero_pd();
 154         fiz2             = _mm_setzero_pd();
 155
 156         /* Reset potential sums */
 157         velecsum         = _mm_setzero_pd();
 158
 159         /* Start inner kernel loop */
 160         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 161         {
 162
 163             /* Get j neighbor index, and coordinate index */
 164             jnrA             = jjnr[jidx];
 165             jnrB             = jjnr[jidx+1];
 166             j_coord_offsetA  = DIM*jnrA;
 167             j_coord_offsetB  = DIM*jnrB;
 168
 169             /* load j atom coordinates */
 170             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 171                                               &jx0,&jy0,&jz0);
 172
 173             /* Calculate displacement vector */
 174             dx00             = _mm_sub_pd(ix0,jx0);
 175             dy00             = _mm_sub_pd(iy0,jy0);
 176             dz00             = _mm_sub_pd(iz0,jz0);
 177             dx10             = _mm_sub_pd(ix1,jx0);
 178             dy10             = _mm_sub_pd(iy1,jy0);
 179             dz10             = _mm_sub_pd(iz1,jz0);
 180             dx20             = _mm_sub_pd(ix2,jx0);
 181             dy20             = _mm_sub_pd(iy2,jy0);
 182             dz20             = _mm_sub_pd(iz2,jz0);
 183
 184             /* Calculate squared distance and things based on it */
 185             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 186             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 187             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 188
 189             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 190             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 191             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 192
 193             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 194             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 195             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 196
 197             /* Load parameters for j particles */
 198             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 199
 200             fjx0             = _mm_setzero_pd();
 201             fjy0             = _mm_setzero_pd();
 202             fjz0             = _mm_setzero_pd();
 203
 204             /**************************
 205              * CALCULATE INTERACTIONS *
 206              **************************/
 207
 208             /* Compute parameters for interactions between i and j atoms */
 209             qq00             = _mm_mul_pd(iq0,jq0);
 210
 211             /* REACTION-FIELD ELECTROSTATICS */
 212             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 213             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 214
 215             /* Update potential sum for this i atom from the interaction with this j atom. */
 216             velecsum         = _mm_add_pd(velecsum,velec);
 217
 218             fscal            = felec;
 219
 220             /* Calculate temporary vectorial force */
 221             tx               = _mm_mul_pd(fscal,dx00);
 222             ty               = _mm_mul_pd(fscal,dy00);
 223             tz               = _mm_mul_pd(fscal,dz00);
 224
 225             /* Update vectorial force */
 226             fix0             = _mm_add_pd(fix0,tx);
 227             fiy0             = _mm_add_pd(fiy0,ty);
 228             fiz0             = _mm_add_pd(fiz0,tz);
 229
 230             fjx0             = _mm_add_pd(fjx0,tx);
 231             fjy0             = _mm_add_pd(fjy0,ty);
 232             fjz0             = _mm_add_pd(fjz0,tz);
 233
 234             /**************************
 235              * CALCULATE INTERACTIONS *
 236              **************************/
 237
 238             /* Compute parameters for interactions between i and j atoms */
 239             qq10             = _mm_mul_pd(iq1,jq0);
 240
 241             /* REACTION-FIELD ELECTROSTATICS */
 242             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 243             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 244
 245             /* Update potential sum for this i atom from the interaction with this j atom. */
 246             velecsum         = _mm_add_pd(velecsum,velec);
 247
 248             fscal            = felec;
 249
 250             /* Calculate temporary vectorial force */
 251             tx               = _mm_mul_pd(fscal,dx10);
 252             ty               = _mm_mul_pd(fscal,dy10);
 253             tz               = _mm_mul_pd(fscal,dz10);
 254
 255             /* Update vectorial force */
 256             fix1             = _mm_add_pd(fix1,tx);
 257             fiy1             = _mm_add_pd(fiy1,ty);
 258             fiz1             = _mm_add_pd(fiz1,tz);
 259
 260             fjx0             = _mm_add_pd(fjx0,tx);
 261             fjy0             = _mm_add_pd(fjy0,ty);
 262             fjz0             = _mm_add_pd(fjz0,tz);
 263
 264             /**************************
 265              * CALCULATE INTERACTIONS *
 266              **************************/
 267
 268             /* Compute parameters for interactions between i and j atoms */
 269             qq20             = _mm_mul_pd(iq2,jq0);
 270
 271             /* REACTION-FIELD ELECTROSTATICS */
 272             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 273             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 274
 275             /* Update potential sum for this i atom from the interaction with this j atom. */
 276             velecsum         = _mm_add_pd(velecsum,velec);
 277
 278             fscal            = felec;
 279
 280             /* Calculate temporary vectorial force */
 281             tx               = _mm_mul_pd(fscal,dx20);
 282             ty               = _mm_mul_pd(fscal,dy20);
 283             tz               = _mm_mul_pd(fscal,dz20);
 284
 285             /* Update vectorial force */
 286             fix2             = _mm_add_pd(fix2,tx);
 287             fiy2             = _mm_add_pd(fiy2,ty);
 288             fiz2             = _mm_add_pd(fiz2,tz);
 289
 290             fjx0             = _mm_add_pd(fjx0,tx);
 291             fjy0             = _mm_add_pd(fjy0,ty);
 292             fjz0             = _mm_add_pd(fjz0,tz);
 293
 294             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 295
 296             /* Inner loop uses 99 flops */
 297         }
 298
 299         if(jidx<j_index_end)
 300         {
 301
 302             jnrA             = jjnr[jidx];
 303             j_coord_offsetA  = DIM*jnrA;
 304
 305             /* load j atom coordinates */
 306             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 307                                               &jx0,&jy0,&jz0);
 308
 309             /* Calculate displacement vector */
 310             dx00             = _mm_sub_pd(ix0,jx0);
 311             dy00             = _mm_sub_pd(iy0,jy0);
 312             dz00             = _mm_sub_pd(iz0,jz0);
 313             dx10             = _mm_sub_pd(ix1,jx0);
 314             dy10             = _mm_sub_pd(iy1,jy0);
 315             dz10             = _mm_sub_pd(iz1,jz0);
 316             dx20             = _mm_sub_pd(ix2,jx0);
 317             dy20             = _mm_sub_pd(iy2,jy0);
 318             dz20             = _mm_sub_pd(iz2,jz0);
 319
 320             /* Calculate squared distance and things based on it */
 321             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 322             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 323             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 324
 325             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 326             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 327             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 328
 329             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 330             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 331             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 332
 333             /* Load parameters for j particles */
 334             jq0              = _mm_load_sd(charge+jnrA+0);
 335
 336             fjx0             = _mm_setzero_pd();
 337             fjy0             = _mm_setzero_pd();
 338             fjz0             = _mm_setzero_pd();
 339
 340             /**************************
 341              * CALCULATE INTERACTIONS *
 342              **************************/
 343
 344             /* Compute parameters for interactions between i and j atoms */
 345             qq00             = _mm_mul_pd(iq0,jq0);
 346
 347             /* REACTION-FIELD ELECTROSTATICS */
 348             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 349             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 350
 351             /* Update potential sum for this i atom from the interaction with this j atom. */
 352             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 353             velecsum         = _mm_add_pd(velecsum,velec);
 354
 355             fscal            = felec;
 356
 357             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 358
 359             /* Calculate temporary vectorial force */
 360             tx               = _mm_mul_pd(fscal,dx00);
 361             ty               = _mm_mul_pd(fscal,dy00);
 362             tz               = _mm_mul_pd(fscal,dz00);
 363
 364             /* Update vectorial force */
 365             fix0             = _mm_add_pd(fix0,tx);
 366             fiy0             = _mm_add_pd(fiy0,ty);
 367             fiz0             = _mm_add_pd(fiz0,tz);
 368
 369             fjx0             = _mm_add_pd(fjx0,tx);
 370             fjy0             = _mm_add_pd(fjy0,ty);
 371             fjz0             = _mm_add_pd(fjz0,tz);
 372
 373             /**************************
 374              * CALCULATE INTERACTIONS *
 375              **************************/
 376
 377             /* Compute parameters for interactions between i and j atoms */
 378             qq10             = _mm_mul_pd(iq1,jq0);
 379
 380             /* REACTION-FIELD ELECTROSTATICS */
 381             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 382             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 383
 384             /* Update potential sum for this i atom from the interaction with this j atom. */
 385             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 386             velecsum         = _mm_add_pd(velecsum,velec);
 387
 388             fscal            = felec;
 389
 390             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 391
 392             /* Calculate temporary vectorial force */
 393             tx               = _mm_mul_pd(fscal,dx10);
 394             ty               = _mm_mul_pd(fscal,dy10);
 395             tz               = _mm_mul_pd(fscal,dz10);
 396
 397             /* Update vectorial force */
 398             fix1             = _mm_add_pd(fix1,tx);
 399             fiy1             = _mm_add_pd(fiy1,ty);
 400             fiz1             = _mm_add_pd(fiz1,tz);
 401
 402             fjx0             = _mm_add_pd(fjx0,tx);
 403             fjy0             = _mm_add_pd(fjy0,ty);
 404             fjz0             = _mm_add_pd(fjz0,tz);
 405
 406             /**************************
 407              * CALCULATE INTERACTIONS *
 408              **************************/
 409
 410             /* Compute parameters for interactions between i and j atoms */
 411             qq20             = _mm_mul_pd(iq2,jq0);
 412
 413             /* REACTION-FIELD ELECTROSTATICS */
 414             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 415             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 416
 417             /* Update potential sum for this i atom from the interaction with this j atom. */
 418             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 419             velecsum         = _mm_add_pd(velecsum,velec);
 420
 421             fscal            = felec;
 422
 423             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 424
 425             /* Calculate temporary vectorial force */
 426             tx               = _mm_mul_pd(fscal,dx20);
 427             ty               = _mm_mul_pd(fscal,dy20);
 428             tz               = _mm_mul_pd(fscal,dz20);
 429
 430             /* Update vectorial force */
 431             fix2             = _mm_add_pd(fix2,tx);
 432             fiy2             = _mm_add_pd(fiy2,ty);
 433             fiz2             = _mm_add_pd(fiz2,tz);
 434
 435             fjx0             = _mm_add_pd(fjx0,tx);
 436             fjy0             = _mm_add_pd(fjy0,ty);
 437             fjz0             = _mm_add_pd(fjz0,tz);
 438
 439             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 440
 441             /* Inner loop uses 99 flops */
 442         }
 443
 444         /* End of innermost loop */
 445
 446         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 447                                               f+i_coord_offset,fshift+i_shift_offset);
 448
 449         ggid                        = gid[iidx];
 450         /* Update potential energies */
 451         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 452
 453         /* Increment number of inner iterations */
 454         inneriter                  += j_index_end - j_index_start;
 455
 456         /* Outer loop uses 19 flops */
 457     }
 458
 459     /* Increment number of outer iterations */
 460     outeriter        += nri;
 461
 462     /* Update outer/inner flops */
 463
 464     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*99);
 465 }
 466 /*
 467  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomW3P1_F_sse2_double
 468  * Electrostatics interaction: ReactionField
 469  * VdW interaction:            None
 470  * Geometry:                   Water3-Particle
 471  * Calculate force/pot:        Force
 472  */
 473 void
 474 nb_kernel_ElecRF_VdwNone_GeomW3P1_F_sse2_double
 475                     (t_nblist                    * gmx_restrict       nlist,
 476                      rvec                        * gmx_restrict          xx,
 477                      rvec                        * gmx_restrict          ff,
 478                      t_forcerec                  * gmx_restrict          fr,
 479                      t_mdatoms                   * gmx_restrict     mdatoms,
 480                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 481                      t_nrnb                      * gmx_restrict        nrnb)
 482 {
 483     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 484      * just 0 for non-waters.
 485      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 486      * jnr indices corresponding to data put in the four positions in the SIMD register.
 487      */
 488     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 489     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 490     int              jnrA,jnrB;
 491     int              j_coord_offsetA,j_coord_offsetB;
 492     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 493     real             rcutoff_scalar;
 494     real             *shiftvec,*fshift,*x,*f;
 495     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 496     int              vdwioffset0;
 497     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 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              vdwjidx0A,vdwjidx0B;
 503     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 504     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 505     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 506     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 507     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 508     real             *charge;
 509     __m128d          dummy_mask,cutoff_mask;
 510     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 511     __m128d          one     = _mm_set1_pd(1.0);
 512     __m128d          two     = _mm_set1_pd(2.0);
 513     x                = xx[0];
 514     f                = ff[0];
 515
 516     nri              = nlist->nri;
 517     iinr             = nlist->iinr;
 518     jindex           = nlist->jindex;
 519     jjnr             = nlist->jjnr;
 520     shiftidx         = nlist->shift;
 521     gid              = nlist->gid;
 522     shiftvec         = fr->shift_vec[0];
 523     fshift           = fr->fshift[0];
 524     facel            = _mm_set1_pd(fr->epsfac);
 525     charge           = mdatoms->chargeA;
 526     krf              = _mm_set1_pd(fr->ic->k_rf);
 527     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 528     crf              = _mm_set1_pd(fr->ic->c_rf);
 529
 530     /* Setup water-specific parameters */
 531     inr              = nlist->iinr[0];
 532     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 533     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 534     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 535
 536     /* Avoid stupid compiler warnings */
 537     jnrA = jnrB = 0;
 538     j_coord_offsetA = 0;
 539     j_coord_offsetB = 0;
 540
 541     outeriter        = 0;
 542     inneriter        = 0;
 543
 544     /* Start outer loop over neighborlists */
 545     for(iidx=0; iidx<nri; iidx++)
 546     {
 547         /* Load shift vector for this list */
 548         i_shift_offset   = DIM*shiftidx[iidx];
 549
 550         /* Load limits for loop over neighbors */
 551         j_index_start    = jindex[iidx];
 552         j_index_end      = jindex[iidx+1];
 553
 554         /* Get outer coordinate index */
 555         inr              = iinr[iidx];
 556         i_coord_offset   = DIM*inr;
 557
 558         /* Load i particle coords and add shift vector */
 559         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 560                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 561
 562         fix0             = _mm_setzero_pd();
 563         fiy0             = _mm_setzero_pd();
 564         fiz0             = _mm_setzero_pd();
 565         fix1             = _mm_setzero_pd();
 566         fiy1             = _mm_setzero_pd();
 567         fiz1             = _mm_setzero_pd();
 568         fix2             = _mm_setzero_pd();
 569         fiy2             = _mm_setzero_pd();
 570         fiz2             = _mm_setzero_pd();
 571
 572         /* Start inner kernel loop */
 573         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 574         {
 575
 576             /* Get j neighbor index, and coordinate index */
 577             jnrA             = jjnr[jidx];
 578             jnrB             = jjnr[jidx+1];
 579             j_coord_offsetA  = DIM*jnrA;
 580             j_coord_offsetB  = DIM*jnrB;
 581
 582             /* load j atom coordinates */
 583             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 584                                               &jx0,&jy0,&jz0);
 585
 586             /* Calculate displacement vector */
 587             dx00             = _mm_sub_pd(ix0,jx0);
 588             dy00             = _mm_sub_pd(iy0,jy0);
 589             dz00             = _mm_sub_pd(iz0,jz0);
 590             dx10             = _mm_sub_pd(ix1,jx0);
 591             dy10             = _mm_sub_pd(iy1,jy0);
 592             dz10             = _mm_sub_pd(iz1,jz0);
 593             dx20             = _mm_sub_pd(ix2,jx0);
 594             dy20             = _mm_sub_pd(iy2,jy0);
 595             dz20             = _mm_sub_pd(iz2,jz0);
 596
 597             /* Calculate squared distance and things based on it */
 598             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 599             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 600             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 601
 602             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 603             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 604             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 605
 606             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 607             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 608             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 609
 610             /* Load parameters for j particles */
 611             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 612
 613             fjx0             = _mm_setzero_pd();
 614             fjy0             = _mm_setzero_pd();
 615             fjz0             = _mm_setzero_pd();
 616
 617             /**************************
 618              * CALCULATE INTERACTIONS *
 619              **************************/
 620
 621             /* Compute parameters for interactions between i and j atoms */
 622             qq00             = _mm_mul_pd(iq0,jq0);
 623
 624             /* REACTION-FIELD ELECTROSTATICS */
 625             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 626
 627             fscal            = felec;
 628
 629             /* Calculate temporary vectorial force */
 630             tx               = _mm_mul_pd(fscal,dx00);
 631             ty               = _mm_mul_pd(fscal,dy00);
 632             tz               = _mm_mul_pd(fscal,dz00);
 633
 634             /* Update vectorial force */
 635             fix0             = _mm_add_pd(fix0,tx);
 636             fiy0             = _mm_add_pd(fiy0,ty);
 637             fiz0             = _mm_add_pd(fiz0,tz);
 638
 639             fjx0             = _mm_add_pd(fjx0,tx);
 640             fjy0             = _mm_add_pd(fjy0,ty);
 641             fjz0             = _mm_add_pd(fjz0,tz);
 642
 643             /**************************
 644              * CALCULATE INTERACTIONS *
 645              **************************/
 646
 647             /* Compute parameters for interactions between i and j atoms */
 648             qq10             = _mm_mul_pd(iq1,jq0);
 649
 650             /* REACTION-FIELD ELECTROSTATICS */
 651             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 652
 653             fscal            = felec;
 654
 655             /* Calculate temporary vectorial force */
 656             tx               = _mm_mul_pd(fscal,dx10);
 657             ty               = _mm_mul_pd(fscal,dy10);
 658             tz               = _mm_mul_pd(fscal,dz10);
 659
 660             /* Update vectorial force */
 661             fix1             = _mm_add_pd(fix1,tx);
 662             fiy1             = _mm_add_pd(fiy1,ty);
 663             fiz1             = _mm_add_pd(fiz1,tz);
 664
 665             fjx0             = _mm_add_pd(fjx0,tx);
 666             fjy0             = _mm_add_pd(fjy0,ty);
 667             fjz0             = _mm_add_pd(fjz0,tz);
 668
 669             /**************************
 670              * CALCULATE INTERACTIONS *
 671              **************************/
 672
 673             /* Compute parameters for interactions between i and j atoms */
 674             qq20             = _mm_mul_pd(iq2,jq0);
 675
 676             /* REACTION-FIELD ELECTROSTATICS */
 677             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 678
 679             fscal            = felec;
 680
 681             /* Calculate temporary vectorial force */
 682             tx               = _mm_mul_pd(fscal,dx20);
 683             ty               = _mm_mul_pd(fscal,dy20);
 684             tz               = _mm_mul_pd(fscal,dz20);
 685
 686             /* Update vectorial force */
 687             fix2             = _mm_add_pd(fix2,tx);
 688             fiy2             = _mm_add_pd(fiy2,ty);
 689             fiz2             = _mm_add_pd(fiz2,tz);
 690
 691             fjx0             = _mm_add_pd(fjx0,tx);
 692             fjy0             = _mm_add_pd(fjy0,ty);
 693             fjz0             = _mm_add_pd(fjz0,tz);
 694
 695             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 696
 697             /* Inner loop uses 84 flops */
 698         }
 699
 700         if(jidx<j_index_end)
 701         {
 702
 703             jnrA             = jjnr[jidx];
 704             j_coord_offsetA  = DIM*jnrA;
 705
 706             /* load j atom coordinates */
 707             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 708                                               &jx0,&jy0,&jz0);
 709
 710             /* Calculate displacement vector */
 711             dx00             = _mm_sub_pd(ix0,jx0);
 712             dy00             = _mm_sub_pd(iy0,jy0);
 713             dz00             = _mm_sub_pd(iz0,jz0);
 714             dx10             = _mm_sub_pd(ix1,jx0);
 715             dy10             = _mm_sub_pd(iy1,jy0);
 716             dz10             = _mm_sub_pd(iz1,jz0);
 717             dx20             = _mm_sub_pd(ix2,jx0);
 718             dy20             = _mm_sub_pd(iy2,jy0);
 719             dz20             = _mm_sub_pd(iz2,jz0);
 720
 721             /* Calculate squared distance and things based on it */
 722             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 723             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 724             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 725
 726             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 727             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 728             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 729
 730             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 731             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 732             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 733
 734             /* Load parameters for j particles */
 735             jq0              = _mm_load_sd(charge+jnrA+0);
 736
 737             fjx0             = _mm_setzero_pd();
 738             fjy0             = _mm_setzero_pd();
 739             fjz0             = _mm_setzero_pd();
 740
 741             /**************************
 742              * CALCULATE INTERACTIONS *
 743              **************************/
 744
 745             /* Compute parameters for interactions between i and j atoms */
 746             qq00             = _mm_mul_pd(iq0,jq0);
 747
 748             /* REACTION-FIELD ELECTROSTATICS */
 749             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 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             /* REACTION-FIELD ELECTROSTATICS */
 777             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 778
 779             fscal            = felec;
 780
 781             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 782
 783             /* Calculate temporary vectorial force */
 784             tx               = _mm_mul_pd(fscal,dx10);
 785             ty               = _mm_mul_pd(fscal,dy10);
 786             tz               = _mm_mul_pd(fscal,dz10);
 787
 788             /* Update vectorial force */
 789             fix1             = _mm_add_pd(fix1,tx);
 790             fiy1             = _mm_add_pd(fiy1,ty);
 791             fiz1             = _mm_add_pd(fiz1,tz);
 792
 793             fjx0             = _mm_add_pd(fjx0,tx);
 794             fjy0             = _mm_add_pd(fjy0,ty);
 795             fjz0             = _mm_add_pd(fjz0,tz);
 796
 797             /**************************
 798              * CALCULATE INTERACTIONS *
 799              **************************/
 800
 801             /* Compute parameters for interactions between i and j atoms */
 802             qq20             = _mm_mul_pd(iq2,jq0);
 803
 804             /* REACTION-FIELD ELECTROSTATICS */
 805             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 806
 807             fscal            = felec;
 808
 809             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 810
 811             /* Calculate temporary vectorial force */
 812             tx               = _mm_mul_pd(fscal,dx20);
 813             ty               = _mm_mul_pd(fscal,dy20);
 814             tz               = _mm_mul_pd(fscal,dz20);
 815
 816             /* Update vectorial force */
 817             fix2             = _mm_add_pd(fix2,tx);
 818             fiy2             = _mm_add_pd(fiy2,ty);
 819             fiz2             = _mm_add_pd(fiz2,tz);
 820
 821             fjx0             = _mm_add_pd(fjx0,tx);
 822             fjy0             = _mm_add_pd(fjy0,ty);
 823             fjz0             = _mm_add_pd(fjz0,tz);
 824
 825             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 826
 827             /* Inner loop uses 84 flops */
 828         }
 829
 830         /* End of innermost loop */
 831
 832         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 833                                               f+i_coord_offset,fshift+i_shift_offset);
 834
 835         /* Increment number of inner iterations */
 836         inneriter                  += j_index_end - j_index_start;
 837
 838         /* Outer loop uses 18 flops */
 839     }
 840
 841     /* Increment number of outer iterations */
 842     outeriter        += nri;
 843
 844     /* Update outer/inner flops */
 845
 846     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*84);
 847 }