src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecRF_VdwNone_GeomW4P1_avx_128_fma_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomW4P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            None
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRF_VdwNone_GeomW4P1_VF_avx_128_fma_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwioffset3;
  87     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  91     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  92     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  93     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     __m128d          dummy_mask,cutoff_mask;
  96     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  97     __m128d          one     = _mm_set1_pd(1.0);
  98     __m128d          two     = _mm_set1_pd(2.0);
  99     x                = xx[0];
 100     f                = ff[0];
 101
 102     nri              = nlist->nri;
 103     iinr             = nlist->iinr;
 104     jindex           = nlist->jindex;
 105     jjnr             = nlist->jjnr;
 106     shiftidx         = nlist->shift;
 107     gid              = nlist->gid;
 108     shiftvec         = fr->shift_vec[0];
 109     fshift           = fr->fshift[0];
 110     facel            = _mm_set1_pd(fr->epsfac);
 111     charge           = mdatoms->chargeA;
 112     krf              = _mm_set1_pd(fr->ic->k_rf);
 113     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 114     crf              = _mm_set1_pd(fr->ic->c_rf);
 115
 116     /* Setup water-specific parameters */
 117     inr              = nlist->iinr[0];
 118     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 119     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 120     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 121
 122     /* Avoid stupid compiler warnings */
 123     jnrA = jnrB = 0;
 124     j_coord_offsetA = 0;
 125     j_coord_offsetB = 0;
 126
 127     outeriter        = 0;
 128     inneriter        = 0;
 129
 130     /* Start outer loop over neighborlists */
 131     for(iidx=0; iidx<nri; iidx++)
 132     {
 133         /* Load shift vector for this list */
 134         i_shift_offset   = DIM*shiftidx[iidx];
 135
 136         /* Load limits for loop over neighbors */
 137         j_index_start    = jindex[iidx];
 138         j_index_end      = jindex[iidx+1];
 139
 140         /* Get outer coordinate index */
 141         inr              = iinr[iidx];
 142         i_coord_offset   = DIM*inr;
 143
 144         /* Load i particle coords and add shift vector */
 145         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 146                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 147
 148         fix1             = _mm_setzero_pd();
 149         fiy1             = _mm_setzero_pd();
 150         fiz1             = _mm_setzero_pd();
 151         fix2             = _mm_setzero_pd();
 152         fiy2             = _mm_setzero_pd();
 153         fiz2             = _mm_setzero_pd();
 154         fix3             = _mm_setzero_pd();
 155         fiy3             = _mm_setzero_pd();
 156         fiz3             = _mm_setzero_pd();
 157
 158         /* Reset potential sums */
 159         velecsum         = _mm_setzero_pd();
 160
 161         /* Start inner kernel loop */
 162         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 163         {
 164
 165             /* Get j neighbor index, and coordinate index */
 166             jnrA             = jjnr[jidx];
 167             jnrB             = jjnr[jidx+1];
 168             j_coord_offsetA  = DIM*jnrA;
 169             j_coord_offsetB  = DIM*jnrB;
 170
 171             /* load j atom coordinates */
 172             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 173                                               &jx0,&jy0,&jz0);
 174
 175             /* Calculate displacement vector */
 176             dx10             = _mm_sub_pd(ix1,jx0);
 177             dy10             = _mm_sub_pd(iy1,jy0);
 178             dz10             = _mm_sub_pd(iz1,jz0);
 179             dx20             = _mm_sub_pd(ix2,jx0);
 180             dy20             = _mm_sub_pd(iy2,jy0);
 181             dz20             = _mm_sub_pd(iz2,jz0);
 182             dx30             = _mm_sub_pd(ix3,jx0);
 183             dy30             = _mm_sub_pd(iy3,jy0);
 184             dz30             = _mm_sub_pd(iz3,jz0);
 185
 186             /* Calculate squared distance and things based on it */
 187             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 188             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 189             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 190
 191             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 192             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 193             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 194
 195             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 196             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 197             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 198
 199             /* Load parameters for j particles */
 200             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 201
 202             fjx0             = _mm_setzero_pd();
 203             fjy0             = _mm_setzero_pd();
 204             fjz0             = _mm_setzero_pd();
 205
 206             /**************************
 207              * CALCULATE INTERACTIONS *
 208              **************************/
 209
 210             /* Compute parameters for interactions between i and j atoms */
 211             qq10             = _mm_mul_pd(iq1,jq0);
 212
 213             /* REACTION-FIELD ELECTROSTATICS */
 214             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 215             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 216
 217             /* Update potential sum for this i atom from the interaction with this j atom. */
 218             velecsum         = _mm_add_pd(velecsum,velec);
 219
 220             fscal            = felec;
 221
 222             /* Update vectorial force */
 223             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 224             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 225             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 226
 227             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 228             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 229             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 230
 231             /**************************
 232              * CALCULATE INTERACTIONS *
 233              **************************/
 234
 235             /* Compute parameters for interactions between i and j atoms */
 236             qq20             = _mm_mul_pd(iq2,jq0);
 237
 238             /* REACTION-FIELD ELECTROSTATICS */
 239             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 240             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 241
 242             /* Update potential sum for this i atom from the interaction with this j atom. */
 243             velecsum         = _mm_add_pd(velecsum,velec);
 244
 245             fscal            = felec;
 246
 247             /* Update vectorial force */
 248             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 249             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 250             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 251
 252             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 253             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 254             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 255
 256             /**************************
 257              * CALCULATE INTERACTIONS *
 258              **************************/
 259
 260             /* Compute parameters for interactions between i and j atoms */
 261             qq30             = _mm_mul_pd(iq3,jq0);
 262
 263             /* REACTION-FIELD ELECTROSTATICS */
 264             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_macc_pd(krf,rsq30,rinv30),crf));
 265             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 266
 267             /* Update potential sum for this i atom from the interaction with this j atom. */
 268             velecsum         = _mm_add_pd(velecsum,velec);
 269
 270             fscal            = felec;
 271
 272             /* Update vectorial force */
 273             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 274             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 275             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 276
 277             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 278             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 279             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 280
 281             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 282
 283             /* Inner loop uses 108 flops */
 284         }
 285
 286         if(jidx<j_index_end)
 287         {
 288
 289             jnrA             = jjnr[jidx];
 290             j_coord_offsetA  = DIM*jnrA;
 291
 292             /* load j atom coordinates */
 293             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 294                                               &jx0,&jy0,&jz0);
 295
 296             /* Calculate displacement vector */
 297             dx10             = _mm_sub_pd(ix1,jx0);
 298             dy10             = _mm_sub_pd(iy1,jy0);
 299             dz10             = _mm_sub_pd(iz1,jz0);
 300             dx20             = _mm_sub_pd(ix2,jx0);
 301             dy20             = _mm_sub_pd(iy2,jy0);
 302             dz20             = _mm_sub_pd(iz2,jz0);
 303             dx30             = _mm_sub_pd(ix3,jx0);
 304             dy30             = _mm_sub_pd(iy3,jy0);
 305             dz30             = _mm_sub_pd(iz3,jz0);
 306
 307             /* Calculate squared distance and things based on it */
 308             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 309             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 310             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 311
 312             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 313             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 314             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 315
 316             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 317             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 318             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 319
 320             /* Load parameters for j particles */
 321             jq0              = _mm_load_sd(charge+jnrA+0);
 322
 323             fjx0             = _mm_setzero_pd();
 324             fjy0             = _mm_setzero_pd();
 325             fjz0             = _mm_setzero_pd();
 326
 327             /**************************
 328              * CALCULATE INTERACTIONS *
 329              **************************/
 330
 331             /* Compute parameters for interactions between i and j atoms */
 332             qq10             = _mm_mul_pd(iq1,jq0);
 333
 334             /* REACTION-FIELD ELECTROSTATICS */
 335             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 336             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 337
 338             /* Update potential sum for this i atom from the interaction with this j atom. */
 339             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 340             velecsum         = _mm_add_pd(velecsum,velec);
 341
 342             fscal            = felec;
 343
 344             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 345
 346             /* Update vectorial force */
 347             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 348             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 349             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 350
 351             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 352             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 353             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 354
 355             /**************************
 356              * CALCULATE INTERACTIONS *
 357              **************************/
 358
 359             /* Compute parameters for interactions between i and j atoms */
 360             qq20             = _mm_mul_pd(iq2,jq0);
 361
 362             /* REACTION-FIELD ELECTROSTATICS */
 363             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 364             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 365
 366             /* Update potential sum for this i atom from the interaction with this j atom. */
 367             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 368             velecsum         = _mm_add_pd(velecsum,velec);
 369
 370             fscal            = felec;
 371
 372             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 373
 374             /* Update vectorial force */
 375             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 376             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 377             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 378
 379             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 380             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 381             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 382
 383             /**************************
 384              * CALCULATE INTERACTIONS *
 385              **************************/
 386
 387             /* Compute parameters for interactions between i and j atoms */
 388             qq30             = _mm_mul_pd(iq3,jq0);
 389
 390             /* REACTION-FIELD ELECTROSTATICS */
 391             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_macc_pd(krf,rsq30,rinv30),crf));
 392             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 393
 394             /* Update potential sum for this i atom from the interaction with this j atom. */
 395             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 396             velecsum         = _mm_add_pd(velecsum,velec);
 397
 398             fscal            = felec;
 399
 400             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 401
 402             /* Update vectorial force */
 403             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 404             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 405             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 406
 407             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 408             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 409             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 410
 411             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 412
 413             /* Inner loop uses 108 flops */
 414         }
 415
 416         /* End of innermost loop */
 417
 418         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 419                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 420
 421         ggid                        = gid[iidx];
 422         /* Update potential energies */
 423         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 424
 425         /* Increment number of inner iterations */
 426         inneriter                  += j_index_end - j_index_start;
 427
 428         /* Outer loop uses 19 flops */
 429     }
 430
 431     /* Increment number of outer iterations */
 432     outeriter        += nri;
 433
 434     /* Update outer/inner flops */
 435
 436     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*108);
 437 }
 438 /*
 439  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomW4P1_F_avx_128_fma_double
 440  * Electrostatics interaction: ReactionField
 441  * VdW interaction:            None
 442  * Geometry:                   Water4-Particle
 443  * Calculate force/pot:        Force
 444  */
 445 void
 446 nb_kernel_ElecRF_VdwNone_GeomW4P1_F_avx_128_fma_double
 447                     (t_nblist                    * gmx_restrict       nlist,
 448                      rvec                        * gmx_restrict          xx,
 449                      rvec                        * gmx_restrict          ff,
 450                      t_forcerec                  * gmx_restrict          fr,
 451                      t_mdatoms                   * gmx_restrict     mdatoms,
 452                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 453                      t_nrnb                      * gmx_restrict        nrnb)
 454 {
 455     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 456      * just 0 for non-waters.
 457      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 458      * jnr indices corresponding to data put in the four positions in the SIMD register.
 459      */
 460     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 461     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 462     int              jnrA,jnrB;
 463     int              j_coord_offsetA,j_coord_offsetB;
 464     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 465     real             rcutoff_scalar;
 466     real             *shiftvec,*fshift,*x,*f;
 467     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 468     int              vdwioffset1;
 469     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 470     int              vdwioffset2;
 471     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 472     int              vdwioffset3;
 473     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 474     int              vdwjidx0A,vdwjidx0B;
 475     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 476     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 477     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 478     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 479     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 480     real             *charge;
 481     __m128d          dummy_mask,cutoff_mask;
 482     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 483     __m128d          one     = _mm_set1_pd(1.0);
 484     __m128d          two     = _mm_set1_pd(2.0);
 485     x                = xx[0];
 486     f                = ff[0];
 487
 488     nri              = nlist->nri;
 489     iinr             = nlist->iinr;
 490     jindex           = nlist->jindex;
 491     jjnr             = nlist->jjnr;
 492     shiftidx         = nlist->shift;
 493     gid              = nlist->gid;
 494     shiftvec         = fr->shift_vec[0];
 495     fshift           = fr->fshift[0];
 496     facel            = _mm_set1_pd(fr->epsfac);
 497     charge           = mdatoms->chargeA;
 498     krf              = _mm_set1_pd(fr->ic->k_rf);
 499     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 500     crf              = _mm_set1_pd(fr->ic->c_rf);
 501
 502     /* Setup water-specific parameters */
 503     inr              = nlist->iinr[0];
 504     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 505     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 506     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 507
 508     /* Avoid stupid compiler warnings */
 509     jnrA = jnrB = 0;
 510     j_coord_offsetA = 0;
 511     j_coord_offsetB = 0;
 512
 513     outeriter        = 0;
 514     inneriter        = 0;
 515
 516     /* Start outer loop over neighborlists */
 517     for(iidx=0; iidx<nri; iidx++)
 518     {
 519         /* Load shift vector for this list */
 520         i_shift_offset   = DIM*shiftidx[iidx];
 521
 522         /* Load limits for loop over neighbors */
 523         j_index_start    = jindex[iidx];
 524         j_index_end      = jindex[iidx+1];
 525
 526         /* Get outer coordinate index */
 527         inr              = iinr[iidx];
 528         i_coord_offset   = DIM*inr;
 529
 530         /* Load i particle coords and add shift vector */
 531         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 532                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 533
 534         fix1             = _mm_setzero_pd();
 535         fiy1             = _mm_setzero_pd();
 536         fiz1             = _mm_setzero_pd();
 537         fix2             = _mm_setzero_pd();
 538         fiy2             = _mm_setzero_pd();
 539         fiz2             = _mm_setzero_pd();
 540         fix3             = _mm_setzero_pd();
 541         fiy3             = _mm_setzero_pd();
 542         fiz3             = _mm_setzero_pd();
 543
 544         /* Start inner kernel loop */
 545         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 546         {
 547
 548             /* Get j neighbor index, and coordinate index */
 549             jnrA             = jjnr[jidx];
 550             jnrB             = jjnr[jidx+1];
 551             j_coord_offsetA  = DIM*jnrA;
 552             j_coord_offsetB  = DIM*jnrB;
 553
 554             /* load j atom coordinates */
 555             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 556                                               &jx0,&jy0,&jz0);
 557
 558             /* Calculate displacement vector */
 559             dx10             = _mm_sub_pd(ix1,jx0);
 560             dy10             = _mm_sub_pd(iy1,jy0);
 561             dz10             = _mm_sub_pd(iz1,jz0);
 562             dx20             = _mm_sub_pd(ix2,jx0);
 563             dy20             = _mm_sub_pd(iy2,jy0);
 564             dz20             = _mm_sub_pd(iz2,jz0);
 565             dx30             = _mm_sub_pd(ix3,jx0);
 566             dy30             = _mm_sub_pd(iy3,jy0);
 567             dz30             = _mm_sub_pd(iz3,jz0);
 568
 569             /* Calculate squared distance and things based on it */
 570             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 571             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 572             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 573
 574             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 575             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 576             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 577
 578             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 579             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 580             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 581
 582             /* Load parameters for j particles */
 583             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 584
 585             fjx0             = _mm_setzero_pd();
 586             fjy0             = _mm_setzero_pd();
 587             fjz0             = _mm_setzero_pd();
 588
 589             /**************************
 590              * CALCULATE INTERACTIONS *
 591              **************************/
 592
 593             /* Compute parameters for interactions between i and j atoms */
 594             qq10             = _mm_mul_pd(iq1,jq0);
 595
 596             /* REACTION-FIELD ELECTROSTATICS */
 597             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 598
 599             fscal            = felec;
 600
 601             /* Update vectorial force */
 602             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 603             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 604             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 605
 606             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 607             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 608             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 609
 610             /**************************
 611              * CALCULATE INTERACTIONS *
 612              **************************/
 613
 614             /* Compute parameters for interactions between i and j atoms */
 615             qq20             = _mm_mul_pd(iq2,jq0);
 616
 617             /* REACTION-FIELD ELECTROSTATICS */
 618             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 619
 620             fscal            = felec;
 621
 622             /* Update vectorial force */
 623             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 624             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 625             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 626
 627             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 628             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 629             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 630
 631             /**************************
 632              * CALCULATE INTERACTIONS *
 633              **************************/
 634
 635             /* Compute parameters for interactions between i and j atoms */
 636             qq30             = _mm_mul_pd(iq3,jq0);
 637
 638             /* REACTION-FIELD ELECTROSTATICS */
 639             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 640
 641             fscal            = felec;
 642
 643             /* Update vectorial force */
 644             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 645             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 646             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 647
 648             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 649             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 650             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 651
 652             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 653
 654             /* Inner loop uses 93 flops */
 655         }
 656
 657         if(jidx<j_index_end)
 658         {
 659
 660             jnrA             = jjnr[jidx];
 661             j_coord_offsetA  = DIM*jnrA;
 662
 663             /* load j atom coordinates */
 664             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 665                                               &jx0,&jy0,&jz0);
 666
 667             /* Calculate displacement vector */
 668             dx10             = _mm_sub_pd(ix1,jx0);
 669             dy10             = _mm_sub_pd(iy1,jy0);
 670             dz10             = _mm_sub_pd(iz1,jz0);
 671             dx20             = _mm_sub_pd(ix2,jx0);
 672             dy20             = _mm_sub_pd(iy2,jy0);
 673             dz20             = _mm_sub_pd(iz2,jz0);
 674             dx30             = _mm_sub_pd(ix3,jx0);
 675             dy30             = _mm_sub_pd(iy3,jy0);
 676             dz30             = _mm_sub_pd(iz3,jz0);
 677
 678             /* Calculate squared distance and things based on it */
 679             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 680             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 681             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 682
 683             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 684             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 685             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 686
 687             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 688             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 689             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 690
 691             /* Load parameters for j particles */
 692             jq0              = _mm_load_sd(charge+jnrA+0);
 693
 694             fjx0             = _mm_setzero_pd();
 695             fjy0             = _mm_setzero_pd();
 696             fjz0             = _mm_setzero_pd();
 697
 698             /**************************
 699              * CALCULATE INTERACTIONS *
 700              **************************/
 701
 702             /* Compute parameters for interactions between i and j atoms */
 703             qq10             = _mm_mul_pd(iq1,jq0);
 704
 705             /* REACTION-FIELD ELECTROSTATICS */
 706             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 707
 708             fscal            = felec;
 709
 710             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 711
 712             /* Update vectorial force */
 713             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 714             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 715             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 716
 717             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 718             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 719             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 720
 721             /**************************
 722              * CALCULATE INTERACTIONS *
 723              **************************/
 724
 725             /* Compute parameters for interactions between i and j atoms */
 726             qq20             = _mm_mul_pd(iq2,jq0);
 727
 728             /* REACTION-FIELD ELECTROSTATICS */
 729             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 730
 731             fscal            = felec;
 732
 733             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 734
 735             /* Update vectorial force */
 736             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 737             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 738             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 739
 740             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 741             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 742             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 743
 744             /**************************
 745              * CALCULATE INTERACTIONS *
 746              **************************/
 747
 748             /* Compute parameters for interactions between i and j atoms */
 749             qq30             = _mm_mul_pd(iq3,jq0);
 750
 751             /* REACTION-FIELD ELECTROSTATICS */
 752             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 753
 754             fscal            = felec;
 755
 756             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 757
 758             /* Update vectorial force */
 759             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 760             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 761             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 762
 763             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 764             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 765             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 766
 767             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 768
 769             /* Inner loop uses 93 flops */
 770         }
 771
 772         /* End of innermost loop */
 773
 774         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 775                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 776
 777         /* Increment number of inner iterations */
 778         inneriter                  += j_index_end - j_index_start;
 779
 780         /* Outer loop uses 18 flops */
 781     }
 782
 783     /* Increment number of outer iterations */
 784     outeriter        += nri;
 785
 786     /* Update outer/inner flops */
 787
 788     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*93);
 789 }