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) 2012,2013,2014, 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 "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/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_ElecRF_VdwNone_GeomW4P1_VF_avx_128_fma_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            None
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRF_VdwNone_GeomW4P1_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              vdwioffset1;
  81     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  82     int              vdwioffset2;
  83     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  84     int              vdwioffset3;
  85     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  86     int              vdwjidx0A,vdwjidx0B;
  87     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  89     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  90     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  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     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 117     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 118     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 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+DIM,
 144                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 145
 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         fix3             = _mm_setzero_pd();
 153         fiy3             = _mm_setzero_pd();
 154         fiz3             = _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             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             dx30             = _mm_sub_pd(ix3,jx0);
 181             dy30             = _mm_sub_pd(iy3,jy0);
 182             dz30             = _mm_sub_pd(iz3,jz0);
 183
 184             /* Calculate squared distance and things based on it */
 185             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 186             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 187             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 188
 189             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 190             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 191             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 192
 193             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 194             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 195             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 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             qq10             = _mm_mul_pd(iq1,jq0);
 210
 211             /* REACTION-FIELD ELECTROSTATICS */
 212             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 213             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,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             /* Update vectorial force */
 221             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 222             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 223             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 224
 225             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 226             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 227             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 228
 229             /**************************
 230              * CALCULATE INTERACTIONS *
 231              **************************/
 232
 233             /* Compute parameters for interactions between i and j atoms */
 234             qq20             = _mm_mul_pd(iq2,jq0);
 235
 236             /* REACTION-FIELD ELECTROSTATICS */
 237             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 238             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 239
 240             /* Update potential sum for this i atom from the interaction with this j atom. */
 241             velecsum         = _mm_add_pd(velecsum,velec);
 242
 243             fscal            = felec;
 244
 245             /* Update vectorial force */
 246             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 247             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 248             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 249
 250             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 251             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 252             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 253
 254             /**************************
 255              * CALCULATE INTERACTIONS *
 256              **************************/
 257
 258             /* Compute parameters for interactions between i and j atoms */
 259             qq30             = _mm_mul_pd(iq3,jq0);
 260
 261             /* REACTION-FIELD ELECTROSTATICS */
 262             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_macc_pd(krf,rsq30,rinv30),crf));
 263             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 264
 265             /* Update potential sum for this i atom from the interaction with this j atom. */
 266             velecsum         = _mm_add_pd(velecsum,velec);
 267
 268             fscal            = felec;
 269
 270             /* Update vectorial force */
 271             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 272             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 273             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 274
 275             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 276             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 277             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 278
 279             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 280
 281             /* Inner loop uses 108 flops */
 282         }
 283
 284         if(jidx<j_index_end)
 285         {
 286
 287             jnrA             = jjnr[jidx];
 288             j_coord_offsetA  = DIM*jnrA;
 289
 290             /* load j atom coordinates */
 291             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 292                                               &jx0,&jy0,&jz0);
 293
 294             /* Calculate displacement vector */
 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             dx30             = _mm_sub_pd(ix3,jx0);
 302             dy30             = _mm_sub_pd(iy3,jy0);
 303             dz30             = _mm_sub_pd(iz3,jz0);
 304
 305             /* Calculate squared distance and things based on it */
 306             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 307             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 308             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 309
 310             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 311             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 312             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 313
 314             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 315             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 316             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 317
 318             /* Load parameters for j particles */
 319             jq0              = _mm_load_sd(charge+jnrA+0);
 320
 321             fjx0             = _mm_setzero_pd();
 322             fjy0             = _mm_setzero_pd();
 323             fjz0             = _mm_setzero_pd();
 324
 325             /**************************
 326              * CALCULATE INTERACTIONS *
 327              **************************/
 328
 329             /* Compute parameters for interactions between i and j atoms */
 330             qq10             = _mm_mul_pd(iq1,jq0);
 331
 332             /* REACTION-FIELD ELECTROSTATICS */
 333             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 334             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 335
 336             /* Update potential sum for this i atom from the interaction with this j atom. */
 337             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 338             velecsum         = _mm_add_pd(velecsum,velec);
 339
 340             fscal            = felec;
 341
 342             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 343
 344             /* Update vectorial force */
 345             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 346             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 347             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 348
 349             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 350             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 351             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 352
 353             /**************************
 354              * CALCULATE INTERACTIONS *
 355              **************************/
 356
 357             /* Compute parameters for interactions between i and j atoms */
 358             qq20             = _mm_mul_pd(iq2,jq0);
 359
 360             /* REACTION-FIELD ELECTROSTATICS */
 361             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 362             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 363
 364             /* Update potential sum for this i atom from the interaction with this j atom. */
 365             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 366             velecsum         = _mm_add_pd(velecsum,velec);
 367
 368             fscal            = felec;
 369
 370             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 371
 372             /* Update vectorial force */
 373             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 374             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 375             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 376
 377             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 378             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 379             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 380
 381             /**************************
 382              * CALCULATE INTERACTIONS *
 383              **************************/
 384
 385             /* Compute parameters for interactions between i and j atoms */
 386             qq30             = _mm_mul_pd(iq3,jq0);
 387
 388             /* REACTION-FIELD ELECTROSTATICS */
 389             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_macc_pd(krf,rsq30,rinv30),crf));
 390             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 391
 392             /* Update potential sum for this i atom from the interaction with this j atom. */
 393             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 394             velecsum         = _mm_add_pd(velecsum,velec);
 395
 396             fscal            = felec;
 397
 398             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 399
 400             /* Update vectorial force */
 401             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 402             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 403             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 404
 405             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 406             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 407             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 408
 409             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 410
 411             /* Inner loop uses 108 flops */
 412         }
 413
 414         /* End of innermost loop */
 415
 416         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 417                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 418
 419         ggid                        = gid[iidx];
 420         /* Update potential energies */
 421         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 422
 423         /* Increment number of inner iterations */
 424         inneriter                  += j_index_end - j_index_start;
 425
 426         /* Outer loop uses 19 flops */
 427     }
 428
 429     /* Increment number of outer iterations */
 430     outeriter        += nri;
 431
 432     /* Update outer/inner flops */
 433
 434     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*108);
 435 }
 436 /*
 437  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomW4P1_F_avx_128_fma_double
 438  * Electrostatics interaction: ReactionField
 439  * VdW interaction:            None
 440  * Geometry:                   Water4-Particle
 441  * Calculate force/pot:        Force
 442  */
 443 void
 444 nb_kernel_ElecRF_VdwNone_GeomW4P1_F_avx_128_fma_double
 445                     (t_nblist                    * gmx_restrict       nlist,
 446                      rvec                        * gmx_restrict          xx,
 447                      rvec                        * gmx_restrict          ff,
 448                      t_forcerec                  * gmx_restrict          fr,
 449                      t_mdatoms                   * gmx_restrict     mdatoms,
 450                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 451                      t_nrnb                      * gmx_restrict        nrnb)
 452 {
 453     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 454      * just 0 for non-waters.
 455      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 456      * jnr indices corresponding to data put in the four positions in the SIMD register.
 457      */
 458     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 459     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 460     int              jnrA,jnrB;
 461     int              j_coord_offsetA,j_coord_offsetB;
 462     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 463     real             rcutoff_scalar;
 464     real             *shiftvec,*fshift,*x,*f;
 465     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 466     int              vdwioffset1;
 467     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 468     int              vdwioffset2;
 469     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 470     int              vdwioffset3;
 471     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 472     int              vdwjidx0A,vdwjidx0B;
 473     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 474     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 475     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 476     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 477     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 478     real             *charge;
 479     __m128d          dummy_mask,cutoff_mask;
 480     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 481     __m128d          one     = _mm_set1_pd(1.0);
 482     __m128d          two     = _mm_set1_pd(2.0);
 483     x                = xx[0];
 484     f                = ff[0];
 485
 486     nri              = nlist->nri;
 487     iinr             = nlist->iinr;
 488     jindex           = nlist->jindex;
 489     jjnr             = nlist->jjnr;
 490     shiftidx         = nlist->shift;
 491     gid              = nlist->gid;
 492     shiftvec         = fr->shift_vec[0];
 493     fshift           = fr->fshift[0];
 494     facel            = _mm_set1_pd(fr->epsfac);
 495     charge           = mdatoms->chargeA;
 496     krf              = _mm_set1_pd(fr->ic->k_rf);
 497     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 498     crf              = _mm_set1_pd(fr->ic->c_rf);
 499
 500     /* Setup water-specific parameters */
 501     inr              = nlist->iinr[0];
 502     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 503     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 504     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 505
 506     /* Avoid stupid compiler warnings */
 507     jnrA = jnrB = 0;
 508     j_coord_offsetA = 0;
 509     j_coord_offsetB = 0;
 510
 511     outeriter        = 0;
 512     inneriter        = 0;
 513
 514     /* Start outer loop over neighborlists */
 515     for(iidx=0; iidx<nri; iidx++)
 516     {
 517         /* Load shift vector for this list */
 518         i_shift_offset   = DIM*shiftidx[iidx];
 519
 520         /* Load limits for loop over neighbors */
 521         j_index_start    = jindex[iidx];
 522         j_index_end      = jindex[iidx+1];
 523
 524         /* Get outer coordinate index */
 525         inr              = iinr[iidx];
 526         i_coord_offset   = DIM*inr;
 527
 528         /* Load i particle coords and add shift vector */
 529         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 530                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 531
 532         fix1             = _mm_setzero_pd();
 533         fiy1             = _mm_setzero_pd();
 534         fiz1             = _mm_setzero_pd();
 535         fix2             = _mm_setzero_pd();
 536         fiy2             = _mm_setzero_pd();
 537         fiz2             = _mm_setzero_pd();
 538         fix3             = _mm_setzero_pd();
 539         fiy3             = _mm_setzero_pd();
 540         fiz3             = _mm_setzero_pd();
 541
 542         /* Start inner kernel loop */
 543         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 544         {
 545
 546             /* Get j neighbor index, and coordinate index */
 547             jnrA             = jjnr[jidx];
 548             jnrB             = jjnr[jidx+1];
 549             j_coord_offsetA  = DIM*jnrA;
 550             j_coord_offsetB  = DIM*jnrB;
 551
 552             /* load j atom coordinates */
 553             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 554                                               &jx0,&jy0,&jz0);
 555
 556             /* Calculate displacement vector */
 557             dx10             = _mm_sub_pd(ix1,jx0);
 558             dy10             = _mm_sub_pd(iy1,jy0);
 559             dz10             = _mm_sub_pd(iz1,jz0);
 560             dx20             = _mm_sub_pd(ix2,jx0);
 561             dy20             = _mm_sub_pd(iy2,jy0);
 562             dz20             = _mm_sub_pd(iz2,jz0);
 563             dx30             = _mm_sub_pd(ix3,jx0);
 564             dy30             = _mm_sub_pd(iy3,jy0);
 565             dz30             = _mm_sub_pd(iz3,jz0);
 566
 567             /* Calculate squared distance and things based on it */
 568             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 569             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 570             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 571
 572             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 573             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 574             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 575
 576             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 577             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 578             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 579
 580             /* Load parameters for j particles */
 581             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 582
 583             fjx0             = _mm_setzero_pd();
 584             fjy0             = _mm_setzero_pd();
 585             fjz0             = _mm_setzero_pd();
 586
 587             /**************************
 588              * CALCULATE INTERACTIONS *
 589              **************************/
 590
 591             /* Compute parameters for interactions between i and j atoms */
 592             qq10             = _mm_mul_pd(iq1,jq0);
 593
 594             /* REACTION-FIELD ELECTROSTATICS */
 595             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 596
 597             fscal            = felec;
 598
 599             /* Update vectorial force */
 600             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 601             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 602             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 603
 604             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 605             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 606             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 607
 608             /**************************
 609              * CALCULATE INTERACTIONS *
 610              **************************/
 611
 612             /* Compute parameters for interactions between i and j atoms */
 613             qq20             = _mm_mul_pd(iq2,jq0);
 614
 615             /* REACTION-FIELD ELECTROSTATICS */
 616             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 617
 618             fscal            = felec;
 619
 620             /* Update vectorial force */
 621             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 622             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 623             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 624
 625             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 626             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 627             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 628
 629             /**************************
 630              * CALCULATE INTERACTIONS *
 631              **************************/
 632
 633             /* Compute parameters for interactions between i and j atoms */
 634             qq30             = _mm_mul_pd(iq3,jq0);
 635
 636             /* REACTION-FIELD ELECTROSTATICS */
 637             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 638
 639             fscal            = felec;
 640
 641             /* Update vectorial force */
 642             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 643             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 644             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 645
 646             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 647             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 648             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 649
 650             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 651
 652             /* Inner loop uses 93 flops */
 653         }
 654
 655         if(jidx<j_index_end)
 656         {
 657
 658             jnrA             = jjnr[jidx];
 659             j_coord_offsetA  = DIM*jnrA;
 660
 661             /* load j atom coordinates */
 662             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 663                                               &jx0,&jy0,&jz0);
 664
 665             /* Calculate displacement vector */
 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             dx30             = _mm_sub_pd(ix3,jx0);
 673             dy30             = _mm_sub_pd(iy3,jy0);
 674             dz30             = _mm_sub_pd(iz3,jz0);
 675
 676             /* Calculate squared distance and things based on it */
 677             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 678             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 679             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 680
 681             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 682             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 683             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 684
 685             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 686             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 687             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 688
 689             /* Load parameters for j particles */
 690             jq0              = _mm_load_sd(charge+jnrA+0);
 691
 692             fjx0             = _mm_setzero_pd();
 693             fjy0             = _mm_setzero_pd();
 694             fjz0             = _mm_setzero_pd();
 695
 696             /**************************
 697              * CALCULATE INTERACTIONS *
 698              **************************/
 699
 700             /* Compute parameters for interactions between i and j atoms */
 701             qq10             = _mm_mul_pd(iq1,jq0);
 702
 703             /* REACTION-FIELD ELECTROSTATICS */
 704             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 705
 706             fscal            = felec;
 707
 708             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 709
 710             /* Update vectorial force */
 711             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 712             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 713             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 714
 715             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 716             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 717             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 718
 719             /**************************
 720              * CALCULATE INTERACTIONS *
 721              **************************/
 722
 723             /* Compute parameters for interactions between i and j atoms */
 724             qq20             = _mm_mul_pd(iq2,jq0);
 725
 726             /* REACTION-FIELD ELECTROSTATICS */
 727             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 728
 729             fscal            = felec;
 730
 731             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 732
 733             /* Update vectorial force */
 734             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 735             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 736             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 737
 738             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 739             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 740             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 741
 742             /**************************
 743              * CALCULATE INTERACTIONS *
 744              **************************/
 745
 746             /* Compute parameters for interactions between i and j atoms */
 747             qq30             = _mm_mul_pd(iq3,jq0);
 748
 749             /* REACTION-FIELD ELECTROSTATICS */
 750             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 751
 752             fscal            = felec;
 753
 754             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 755
 756             /* Update vectorial force */
 757             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 758             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 759             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 760
 761             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 762             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 763             fjz0             = _mm_macc_pd(dz30,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(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 773                                               f+i_coord_offset+DIM,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_W4_F,outeriter*18 + inneriter*93);
 787 }