src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCoul_VdwNone_GeomW3P1_avx_128_fma_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 avx_128_fma_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_avx_128_fma_double.h"
  48 #include "kernelutil_x86_avx_128_fma_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomW3P1_VF_avx_128_fma_double
  52  * Electrostatics interaction: Coulomb
  53  * VdW interaction:            None
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCoul_VdwNone_GeomW3P1_VF_avx_128_fma_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
 111     /* Setup water-specific parameters */
 112     inr              = nlist->iinr[0];
 113     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 114     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 115     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 116
 117     /* Avoid stupid compiler warnings */
 118     jnrA = jnrB = 0;
 119     j_coord_offsetA = 0;
 120     j_coord_offsetB = 0;
 121
 122     outeriter        = 0;
 123     inneriter        = 0;
 124
 125     /* Start outer loop over neighborlists */
 126     for(iidx=0; iidx<nri; iidx++)
 127     {
 128         /* Load shift vector for this list */
 129         i_shift_offset   = DIM*shiftidx[iidx];
 130
 131         /* Load limits for loop over neighbors */
 132         j_index_start    = jindex[iidx];
 133         j_index_end      = jindex[iidx+1];
 134
 135         /* Get outer coordinate index */
 136         inr              = iinr[iidx];
 137         i_coord_offset   = DIM*inr;
 138
 139         /* Load i particle coords and add shift vector */
 140         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 141                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 142
 143         fix0             = _mm_setzero_pd();
 144         fiy0             = _mm_setzero_pd();
 145         fiz0             = _mm_setzero_pd();
 146         fix1             = _mm_setzero_pd();
 147         fiy1             = _mm_setzero_pd();
 148         fiz1             = _mm_setzero_pd();
 149         fix2             = _mm_setzero_pd();
 150         fiy2             = _mm_setzero_pd();
 151         fiz2             = _mm_setzero_pd();
 152
 153         /* Reset potential sums */
 154         velecsum         = _mm_setzero_pd();
 155
 156         /* Start inner kernel loop */
 157         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 158         {
 159
 160             /* Get j neighbor index, and coordinate index */
 161             jnrA             = jjnr[jidx];
 162             jnrB             = jjnr[jidx+1];
 163             j_coord_offsetA  = DIM*jnrA;
 164             j_coord_offsetB  = DIM*jnrB;
 165
 166             /* load j atom coordinates */
 167             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 168                                               &jx0,&jy0,&jz0);
 169
 170             /* Calculate displacement vector */
 171             dx00             = _mm_sub_pd(ix0,jx0);
 172             dy00             = _mm_sub_pd(iy0,jy0);
 173             dz00             = _mm_sub_pd(iz0,jz0);
 174             dx10             = _mm_sub_pd(ix1,jx0);
 175             dy10             = _mm_sub_pd(iy1,jy0);
 176             dz10             = _mm_sub_pd(iz1,jz0);
 177             dx20             = _mm_sub_pd(ix2,jx0);
 178             dy20             = _mm_sub_pd(iy2,jy0);
 179             dz20             = _mm_sub_pd(iz2,jz0);
 180
 181             /* Calculate squared distance and things based on it */
 182             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 183             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 184             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 185
 186             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 187             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 188             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 189
 190             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 191             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 192             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 193
 194             /* Load parameters for j particles */
 195             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 196
 197             fjx0             = _mm_setzero_pd();
 198             fjy0             = _mm_setzero_pd();
 199             fjz0             = _mm_setzero_pd();
 200
 201             /**************************
 202              * CALCULATE INTERACTIONS *
 203              **************************/
 204
 205             /* Compute parameters for interactions between i and j atoms */
 206             qq00             = _mm_mul_pd(iq0,jq0);
 207
 208             /* COULOMB ELECTROSTATICS */
 209             velec            = _mm_mul_pd(qq00,rinv00);
 210             felec            = _mm_mul_pd(velec,rinvsq00);
 211
 212             /* Update potential sum for this i atom from the interaction with this j atom. */
 213             velecsum         = _mm_add_pd(velecsum,velec);
 214
 215             fscal            = felec;
 216
 217             /* Update vectorial force */
 218             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 219             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 220             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 221
 222             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 223             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 224             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 225
 226             /**************************
 227              * CALCULATE INTERACTIONS *
 228              **************************/
 229
 230             /* Compute parameters for interactions between i and j atoms */
 231             qq10             = _mm_mul_pd(iq1,jq0);
 232
 233             /* COULOMB ELECTROSTATICS */
 234             velec            = _mm_mul_pd(qq10,rinv10);
 235             felec            = _mm_mul_pd(velec,rinvsq10);
 236
 237             /* Update potential sum for this i atom from the interaction with this j atom. */
 238             velecsum         = _mm_add_pd(velecsum,velec);
 239
 240             fscal            = felec;
 241
 242             /* Update vectorial force */
 243             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 244             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 245             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 246
 247             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 248             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 249             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 250
 251             /**************************
 252              * CALCULATE INTERACTIONS *
 253              **************************/
 254
 255             /* Compute parameters for interactions between i and j atoms */
 256             qq20             = _mm_mul_pd(iq2,jq0);
 257
 258             /* COULOMB ELECTROSTATICS */
 259             velec            = _mm_mul_pd(qq20,rinv20);
 260             felec            = _mm_mul_pd(velec,rinvsq20);
 261
 262             /* Update potential sum for this i atom from the interaction with this j atom. */
 263             velecsum         = _mm_add_pd(velecsum,velec);
 264
 265             fscal            = felec;
 266
 267             /* Update vectorial force */
 268             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 269             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 270             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 271
 272             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 273             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 274             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 275
 276             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 277
 278             /* Inner loop uses 96 flops */
 279         }
 280
 281         if(jidx<j_index_end)
 282         {
 283
 284             jnrA             = jjnr[jidx];
 285             j_coord_offsetA  = DIM*jnrA;
 286
 287             /* load j atom coordinates */
 288             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 289                                               &jx0,&jy0,&jz0);
 290
 291             /* Calculate displacement vector */
 292             dx00             = _mm_sub_pd(ix0,jx0);
 293             dy00             = _mm_sub_pd(iy0,jy0);
 294             dz00             = _mm_sub_pd(iz0,jz0);
 295             dx10             = _mm_sub_pd(ix1,jx0);
 296             dy10             = _mm_sub_pd(iy1,jy0);
 297             dz10             = _mm_sub_pd(iz1,jz0);
 298             dx20             = _mm_sub_pd(ix2,jx0);
 299             dy20             = _mm_sub_pd(iy2,jy0);
 300             dz20             = _mm_sub_pd(iz2,jz0);
 301
 302             /* Calculate squared distance and things based on it */
 303             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 304             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 305             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 306
 307             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 308             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 309             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 310
 311             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 312             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 313             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 314
 315             /* Load parameters for j particles */
 316             jq0              = _mm_load_sd(charge+jnrA+0);
 317
 318             fjx0             = _mm_setzero_pd();
 319             fjy0             = _mm_setzero_pd();
 320             fjz0             = _mm_setzero_pd();
 321
 322             /**************************
 323              * CALCULATE INTERACTIONS *
 324              **************************/
 325
 326             /* Compute parameters for interactions between i and j atoms */
 327             qq00             = _mm_mul_pd(iq0,jq0);
 328
 329             /* COULOMB ELECTROSTATICS */
 330             velec            = _mm_mul_pd(qq00,rinv00);
 331             felec            = _mm_mul_pd(velec,rinvsq00);
 332
 333             /* Update potential sum for this i atom from the interaction with this j atom. */
 334             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 335             velecsum         = _mm_add_pd(velecsum,velec);
 336
 337             fscal            = felec;
 338
 339             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 340
 341             /* Update vectorial force */
 342             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 343             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 344             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 345
 346             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 347             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 348             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 349
 350             /**************************
 351              * CALCULATE INTERACTIONS *
 352              **************************/
 353
 354             /* Compute parameters for interactions between i and j atoms */
 355             qq10             = _mm_mul_pd(iq1,jq0);
 356
 357             /* COULOMB ELECTROSTATICS */
 358             velec            = _mm_mul_pd(qq10,rinv10);
 359             felec            = _mm_mul_pd(velec,rinvsq10);
 360
 361             /* Update potential sum for this i atom from the interaction with this j atom. */
 362             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 363             velecsum         = _mm_add_pd(velecsum,velec);
 364
 365             fscal            = felec;
 366
 367             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 368
 369             /* Update vectorial force */
 370             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 371             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 372             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 373
 374             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 375             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 376             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 377
 378             /**************************
 379              * CALCULATE INTERACTIONS *
 380              **************************/
 381
 382             /* Compute parameters for interactions between i and j atoms */
 383             qq20             = _mm_mul_pd(iq2,jq0);
 384
 385             /* COULOMB ELECTROSTATICS */
 386             velec            = _mm_mul_pd(qq20,rinv20);
 387             felec            = _mm_mul_pd(velec,rinvsq20);
 388
 389             /* Update potential sum for this i atom from the interaction with this j atom. */
 390             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 391             velecsum         = _mm_add_pd(velecsum,velec);
 392
 393             fscal            = felec;
 394
 395             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 396
 397             /* Update vectorial force */
 398             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 399             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 400             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 401
 402             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 403             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 404             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 405
 406             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 407
 408             /* Inner loop uses 96 flops */
 409         }
 410
 411         /* End of innermost loop */
 412
 413         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 414                                               f+i_coord_offset,fshift+i_shift_offset);
 415
 416         ggid                        = gid[iidx];
 417         /* Update potential energies */
 418         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 419
 420         /* Increment number of inner iterations */
 421         inneriter                  += j_index_end - j_index_start;
 422
 423         /* Outer loop uses 19 flops */
 424     }
 425
 426     /* Increment number of outer iterations */
 427     outeriter        += nri;
 428
 429     /* Update outer/inner flops */
 430
 431     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*96);
 432 }
 433 /*
 434  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomW3P1_F_avx_128_fma_double
 435  * Electrostatics interaction: Coulomb
 436  * VdW interaction:            None
 437  * Geometry:                   Water3-Particle
 438  * Calculate force/pot:        Force
 439  */
 440 void
 441 nb_kernel_ElecCoul_VdwNone_GeomW3P1_F_avx_128_fma_double
 442                     (t_nblist                    * gmx_restrict       nlist,
 443                      rvec                        * gmx_restrict          xx,
 444                      rvec                        * gmx_restrict          ff,
 445                      t_forcerec                  * gmx_restrict          fr,
 446                      t_mdatoms                   * gmx_restrict     mdatoms,
 447                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 448                      t_nrnb                      * gmx_restrict        nrnb)
 449 {
 450     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 451      * just 0 for non-waters.
 452      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 453      * jnr indices corresponding to data put in the four positions in the SIMD register.
 454      */
 455     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 456     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 457     int              jnrA,jnrB;
 458     int              j_coord_offsetA,j_coord_offsetB;
 459     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 460     real             rcutoff_scalar;
 461     real             *shiftvec,*fshift,*x,*f;
 462     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 463     int              vdwioffset0;
 464     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 465     int              vdwioffset1;
 466     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 467     int              vdwioffset2;
 468     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 469     int              vdwjidx0A,vdwjidx0B;
 470     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 471     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 472     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 473     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 474     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 475     real             *charge;
 476     __m128d          dummy_mask,cutoff_mask;
 477     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 478     __m128d          one     = _mm_set1_pd(1.0);
 479     __m128d          two     = _mm_set1_pd(2.0);
 480     x                = xx[0];
 481     f                = ff[0];
 482
 483     nri              = nlist->nri;
 484     iinr             = nlist->iinr;
 485     jindex           = nlist->jindex;
 486     jjnr             = nlist->jjnr;
 487     shiftidx         = nlist->shift;
 488     gid              = nlist->gid;
 489     shiftvec         = fr->shift_vec[0];
 490     fshift           = fr->fshift[0];
 491     facel            = _mm_set1_pd(fr->epsfac);
 492     charge           = mdatoms->chargeA;
 493
 494     /* Setup water-specific parameters */
 495     inr              = nlist->iinr[0];
 496     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 497     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 498     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 499
 500     /* Avoid stupid compiler warnings */
 501     jnrA = jnrB = 0;
 502     j_coord_offsetA = 0;
 503     j_coord_offsetB = 0;
 504
 505     outeriter        = 0;
 506     inneriter        = 0;
 507
 508     /* Start outer loop over neighborlists */
 509     for(iidx=0; iidx<nri; iidx++)
 510     {
 511         /* Load shift vector for this list */
 512         i_shift_offset   = DIM*shiftidx[iidx];
 513
 514         /* Load limits for loop over neighbors */
 515         j_index_start    = jindex[iidx];
 516         j_index_end      = jindex[iidx+1];
 517
 518         /* Get outer coordinate index */
 519         inr              = iinr[iidx];
 520         i_coord_offset   = DIM*inr;
 521
 522         /* Load i particle coords and add shift vector */
 523         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 524                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 525
 526         fix0             = _mm_setzero_pd();
 527         fiy0             = _mm_setzero_pd();
 528         fiz0             = _mm_setzero_pd();
 529         fix1             = _mm_setzero_pd();
 530         fiy1             = _mm_setzero_pd();
 531         fiz1             = _mm_setzero_pd();
 532         fix2             = _mm_setzero_pd();
 533         fiy2             = _mm_setzero_pd();
 534         fiz2             = _mm_setzero_pd();
 535
 536         /* Start inner kernel loop */
 537         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 538         {
 539
 540             /* Get j neighbor index, and coordinate index */
 541             jnrA             = jjnr[jidx];
 542             jnrB             = jjnr[jidx+1];
 543             j_coord_offsetA  = DIM*jnrA;
 544             j_coord_offsetB  = DIM*jnrB;
 545
 546             /* load j atom coordinates */
 547             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 548                                               &jx0,&jy0,&jz0);
 549
 550             /* Calculate displacement vector */
 551             dx00             = _mm_sub_pd(ix0,jx0);
 552             dy00             = _mm_sub_pd(iy0,jy0);
 553             dz00             = _mm_sub_pd(iz0,jz0);
 554             dx10             = _mm_sub_pd(ix1,jx0);
 555             dy10             = _mm_sub_pd(iy1,jy0);
 556             dz10             = _mm_sub_pd(iz1,jz0);
 557             dx20             = _mm_sub_pd(ix2,jx0);
 558             dy20             = _mm_sub_pd(iy2,jy0);
 559             dz20             = _mm_sub_pd(iz2,jz0);
 560
 561             /* Calculate squared distance and things based on it */
 562             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 563             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 564             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 565
 566             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 567             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 568             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 569
 570             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 571             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 572             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 573
 574             /* Load parameters for j particles */
 575             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 576
 577             fjx0             = _mm_setzero_pd();
 578             fjy0             = _mm_setzero_pd();
 579             fjz0             = _mm_setzero_pd();
 580
 581             /**************************
 582              * CALCULATE INTERACTIONS *
 583              **************************/
 584
 585             /* Compute parameters for interactions between i and j atoms */
 586             qq00             = _mm_mul_pd(iq0,jq0);
 587
 588             /* COULOMB ELECTROSTATICS */
 589             velec            = _mm_mul_pd(qq00,rinv00);
 590             felec            = _mm_mul_pd(velec,rinvsq00);
 591
 592             fscal            = felec;
 593
 594             /* Update vectorial force */
 595             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 596             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 597             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 598
 599             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 600             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 601             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 602
 603             /**************************
 604              * CALCULATE INTERACTIONS *
 605              **************************/
 606
 607             /* Compute parameters for interactions between i and j atoms */
 608             qq10             = _mm_mul_pd(iq1,jq0);
 609
 610             /* COULOMB ELECTROSTATICS */
 611             velec            = _mm_mul_pd(qq10,rinv10);
 612             felec            = _mm_mul_pd(velec,rinvsq10);
 613
 614             fscal            = felec;
 615
 616             /* Update vectorial force */
 617             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 618             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 619             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 620
 621             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 622             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 623             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 624
 625             /**************************
 626              * CALCULATE INTERACTIONS *
 627              **************************/
 628
 629             /* Compute parameters for interactions between i and j atoms */
 630             qq20             = _mm_mul_pd(iq2,jq0);
 631
 632             /* COULOMB ELECTROSTATICS */
 633             velec            = _mm_mul_pd(qq20,rinv20);
 634             felec            = _mm_mul_pd(velec,rinvsq20);
 635
 636             fscal            = felec;
 637
 638             /* Update vectorial force */
 639             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 640             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 641             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 642
 643             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 644             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 645             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 646
 647             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 648
 649             /* Inner loop uses 93 flops */
 650         }
 651
 652         if(jidx<j_index_end)
 653         {
 654
 655             jnrA             = jjnr[jidx];
 656             j_coord_offsetA  = DIM*jnrA;
 657
 658             /* load j atom coordinates */
 659             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 660                                               &jx0,&jy0,&jz0);
 661
 662             /* Calculate displacement vector */
 663             dx00             = _mm_sub_pd(ix0,jx0);
 664             dy00             = _mm_sub_pd(iy0,jy0);
 665             dz00             = _mm_sub_pd(iz0,jz0);
 666             dx10             = _mm_sub_pd(ix1,jx0);
 667             dy10             = _mm_sub_pd(iy1,jy0);
 668             dz10             = _mm_sub_pd(iz1,jz0);
 669             dx20             = _mm_sub_pd(ix2,jx0);
 670             dy20             = _mm_sub_pd(iy2,jy0);
 671             dz20             = _mm_sub_pd(iz2,jz0);
 672
 673             /* Calculate squared distance and things based on it */
 674             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 675             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 676             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 677
 678             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 679             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 680             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 681
 682             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 683             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 684             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 685
 686             /* Load parameters for j particles */
 687             jq0              = _mm_load_sd(charge+jnrA+0);
 688
 689             fjx0             = _mm_setzero_pd();
 690             fjy0             = _mm_setzero_pd();
 691             fjz0             = _mm_setzero_pd();
 692
 693             /**************************
 694              * CALCULATE INTERACTIONS *
 695              **************************/
 696
 697             /* Compute parameters for interactions between i and j atoms */
 698             qq00             = _mm_mul_pd(iq0,jq0);
 699
 700             /* COULOMB ELECTROSTATICS */
 701             velec            = _mm_mul_pd(qq00,rinv00);
 702             felec            = _mm_mul_pd(velec,rinvsq00);
 703
 704             fscal            = felec;
 705
 706             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 707
 708             /* Update vectorial force */
 709             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 710             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 711             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 712
 713             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 714             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 715             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 716
 717             /**************************
 718              * CALCULATE INTERACTIONS *
 719              **************************/
 720
 721             /* Compute parameters for interactions between i and j atoms */
 722             qq10             = _mm_mul_pd(iq1,jq0);
 723
 724             /* COULOMB ELECTROSTATICS */
 725             velec            = _mm_mul_pd(qq10,rinv10);
 726             felec            = _mm_mul_pd(velec,rinvsq10);
 727
 728             fscal            = felec;
 729
 730             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 731
 732             /* Update vectorial force */
 733             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 734             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 735             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 736
 737             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 738             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 739             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 740
 741             /**************************
 742              * CALCULATE INTERACTIONS *
 743              **************************/
 744
 745             /* Compute parameters for interactions between i and j atoms */
 746             qq20             = _mm_mul_pd(iq2,jq0);
 747
 748             /* COULOMB ELECTROSTATICS */
 749             velec            = _mm_mul_pd(qq20,rinv20);
 750             felec            = _mm_mul_pd(velec,rinvsq20);
 751
 752             fscal            = felec;
 753
 754             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 755
 756             /* Update vectorial force */
 757             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 758             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 759             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 760
 761             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 762             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 763             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 764
 765             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 766
 767             /* Inner loop uses 93 flops */
 768         }
 769
 770         /* End of innermost loop */
 771
 772         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 773                                               f+i_coord_offset,fshift+i_shift_offset);
 774
 775         /* Increment number of inner iterations */
 776         inneriter                  += j_index_end - j_index_start;
 777
 778         /* Outer loop uses 18 flops */
 779     }
 780
 781     /* Increment number of outer iterations */
 782     outeriter        += nri;
 783
 784     /* Update outer/inner flops */
 785
 786     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*93);
 787 }