src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecCoul_VdwNone_GeomW4P1_sse4_1_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse4_1_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomW4P1_VF_sse4_1_double
  51  * Electrostatics interaction: Coulomb
  52  * VdW interaction:            None
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCoul_VdwNone_GeomW4P1_VF_sse4_1_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset1;
  80     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  81     int              vdwioffset2;
  82     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  83     int              vdwioffset3;
  84     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  85     int              vdwjidx0A,vdwjidx0B;
  86     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  88     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  89     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  90     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     __m128d          dummy_mask,cutoff_mask;
  93     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  94     __m128d          one     = _mm_set1_pd(1.0);
  95     __m128d          two     = _mm_set1_pd(2.0);
  96     x                = xx[0];
  97     f                = ff[0];
  98
  99     nri              = nlist->nri;
 100     iinr             = nlist->iinr;
 101     jindex           = nlist->jindex;
 102     jjnr             = nlist->jjnr;
 103     shiftidx         = nlist->shift;
 104     gid              = nlist->gid;
 105     shiftvec         = fr->shift_vec[0];
 106     fshift           = fr->fshift[0];
 107     facel            = _mm_set1_pd(fr->ic->epsfac);
 108     charge           = mdatoms->chargeA;
 109
 110     /* Setup water-specific parameters */
 111     inr              = nlist->iinr[0];
 112     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 113     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 114     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 115
 116     /* Avoid stupid compiler warnings */
 117     jnrA = jnrB = 0;
 118     j_coord_offsetA = 0;
 119     j_coord_offsetB = 0;
 120
 121     outeriter        = 0;
 122     inneriter        = 0;
 123
 124     /* Start outer loop over neighborlists */
 125     for(iidx=0; iidx<nri; iidx++)
 126     {
 127         /* Load shift vector for this list */
 128         i_shift_offset   = DIM*shiftidx[iidx];
 129
 130         /* Load limits for loop over neighbors */
 131         j_index_start    = jindex[iidx];
 132         j_index_end      = jindex[iidx+1];
 133
 134         /* Get outer coordinate index */
 135         inr              = iinr[iidx];
 136         i_coord_offset   = DIM*inr;
 137
 138         /* Load i particle coords and add shift vector */
 139         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 140                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 141
 142         fix1             = _mm_setzero_pd();
 143         fiy1             = _mm_setzero_pd();
 144         fiz1             = _mm_setzero_pd();
 145         fix2             = _mm_setzero_pd();
 146         fiy2             = _mm_setzero_pd();
 147         fiz2             = _mm_setzero_pd();
 148         fix3             = _mm_setzero_pd();
 149         fiy3             = _mm_setzero_pd();
 150         fiz3             = _mm_setzero_pd();
 151
 152         /* Reset potential sums */
 153         velecsum         = _mm_setzero_pd();
 154
 155         /* Start inner kernel loop */
 156         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 157         {
 158
 159             /* Get j neighbor index, and coordinate index */
 160             jnrA             = jjnr[jidx];
 161             jnrB             = jjnr[jidx+1];
 162             j_coord_offsetA  = DIM*jnrA;
 163             j_coord_offsetB  = DIM*jnrB;
 164
 165             /* load j atom coordinates */
 166             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 167                                               &jx0,&jy0,&jz0);
 168
 169             /* Calculate displacement vector */
 170             dx10             = _mm_sub_pd(ix1,jx0);
 171             dy10             = _mm_sub_pd(iy1,jy0);
 172             dz10             = _mm_sub_pd(iz1,jz0);
 173             dx20             = _mm_sub_pd(ix2,jx0);
 174             dy20             = _mm_sub_pd(iy2,jy0);
 175             dz20             = _mm_sub_pd(iz2,jz0);
 176             dx30             = _mm_sub_pd(ix3,jx0);
 177             dy30             = _mm_sub_pd(iy3,jy0);
 178             dz30             = _mm_sub_pd(iz3,jz0);
 179
 180             /* Calculate squared distance and things based on it */
 181             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 182             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 183             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 184
 185             rinv10           = sse41_invsqrt_d(rsq10);
 186             rinv20           = sse41_invsqrt_d(rsq20);
 187             rinv30           = sse41_invsqrt_d(rsq30);
 188
 189             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 190             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 191             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 192
 193             /* Load parameters for j particles */
 194             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 195
 196             fjx0             = _mm_setzero_pd();
 197             fjy0             = _mm_setzero_pd();
 198             fjz0             = _mm_setzero_pd();
 199
 200             /**************************
 201              * CALCULATE INTERACTIONS *
 202              **************************/
 203
 204             /* Compute parameters for interactions between i and j atoms */
 205             qq10             = _mm_mul_pd(iq1,jq0);
 206
 207             /* COULOMB ELECTROSTATICS */
 208             velec            = _mm_mul_pd(qq10,rinv10);
 209             felec            = _mm_mul_pd(velec,rinvsq10);
 210
 211             /* Update potential sum for this i atom from the interaction with this j atom. */
 212             velecsum         = _mm_add_pd(velecsum,velec);
 213
 214             fscal            = felec;
 215
 216             /* Calculate temporary vectorial force */
 217             tx               = _mm_mul_pd(fscal,dx10);
 218             ty               = _mm_mul_pd(fscal,dy10);
 219             tz               = _mm_mul_pd(fscal,dz10);
 220
 221             /* Update vectorial force */
 222             fix1             = _mm_add_pd(fix1,tx);
 223             fiy1             = _mm_add_pd(fiy1,ty);
 224             fiz1             = _mm_add_pd(fiz1,tz);
 225
 226             fjx0             = _mm_add_pd(fjx0,tx);
 227             fjy0             = _mm_add_pd(fjy0,ty);
 228             fjz0             = _mm_add_pd(fjz0,tz);
 229
 230             /**************************
 231              * CALCULATE INTERACTIONS *
 232              **************************/
 233
 234             /* Compute parameters for interactions between i and j atoms */
 235             qq20             = _mm_mul_pd(iq2,jq0);
 236
 237             /* COULOMB ELECTROSTATICS */
 238             velec            = _mm_mul_pd(qq20,rinv20);
 239             felec            = _mm_mul_pd(velec,rinvsq20);
 240
 241             /* Update potential sum for this i atom from the interaction with this j atom. */
 242             velecsum         = _mm_add_pd(velecsum,velec);
 243
 244             fscal            = felec;
 245
 246             /* Calculate temporary vectorial force */
 247             tx               = _mm_mul_pd(fscal,dx20);
 248             ty               = _mm_mul_pd(fscal,dy20);
 249             tz               = _mm_mul_pd(fscal,dz20);
 250
 251             /* Update vectorial force */
 252             fix2             = _mm_add_pd(fix2,tx);
 253             fiy2             = _mm_add_pd(fiy2,ty);
 254             fiz2             = _mm_add_pd(fiz2,tz);
 255
 256             fjx0             = _mm_add_pd(fjx0,tx);
 257             fjy0             = _mm_add_pd(fjy0,ty);
 258             fjz0             = _mm_add_pd(fjz0,tz);
 259
 260             /**************************
 261              * CALCULATE INTERACTIONS *
 262              **************************/
 263
 264             /* Compute parameters for interactions between i and j atoms */
 265             qq30             = _mm_mul_pd(iq3,jq0);
 266
 267             /* COULOMB ELECTROSTATICS */
 268             velec            = _mm_mul_pd(qq30,rinv30);
 269             felec            = _mm_mul_pd(velec,rinvsq30);
 270
 271             /* Update potential sum for this i atom from the interaction with this j atom. */
 272             velecsum         = _mm_add_pd(velecsum,velec);
 273
 274             fscal            = felec;
 275
 276             /* Calculate temporary vectorial force */
 277             tx               = _mm_mul_pd(fscal,dx30);
 278             ty               = _mm_mul_pd(fscal,dy30);
 279             tz               = _mm_mul_pd(fscal,dz30);
 280
 281             /* Update vectorial force */
 282             fix3             = _mm_add_pd(fix3,tx);
 283             fiy3             = _mm_add_pd(fiy3,ty);
 284             fiz3             = _mm_add_pd(fiz3,tz);
 285
 286             fjx0             = _mm_add_pd(fjx0,tx);
 287             fjy0             = _mm_add_pd(fjy0,ty);
 288             fjz0             = _mm_add_pd(fjz0,tz);
 289
 290             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 291
 292             /* Inner loop uses 87 flops */
 293         }
 294
 295         if(jidx<j_index_end)
 296         {
 297
 298             jnrA             = jjnr[jidx];
 299             j_coord_offsetA  = DIM*jnrA;
 300
 301             /* load j atom coordinates */
 302             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 303                                               &jx0,&jy0,&jz0);
 304
 305             /* Calculate displacement vector */
 306             dx10             = _mm_sub_pd(ix1,jx0);
 307             dy10             = _mm_sub_pd(iy1,jy0);
 308             dz10             = _mm_sub_pd(iz1,jz0);
 309             dx20             = _mm_sub_pd(ix2,jx0);
 310             dy20             = _mm_sub_pd(iy2,jy0);
 311             dz20             = _mm_sub_pd(iz2,jz0);
 312             dx30             = _mm_sub_pd(ix3,jx0);
 313             dy30             = _mm_sub_pd(iy3,jy0);
 314             dz30             = _mm_sub_pd(iz3,jz0);
 315
 316             /* Calculate squared distance and things based on it */
 317             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 318             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 319             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 320
 321             rinv10           = sse41_invsqrt_d(rsq10);
 322             rinv20           = sse41_invsqrt_d(rsq20);
 323             rinv30           = sse41_invsqrt_d(rsq30);
 324
 325             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 326             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 327             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 328
 329             /* Load parameters for j particles */
 330             jq0              = _mm_load_sd(charge+jnrA+0);
 331
 332             fjx0             = _mm_setzero_pd();
 333             fjy0             = _mm_setzero_pd();
 334             fjz0             = _mm_setzero_pd();
 335
 336             /**************************
 337              * CALCULATE INTERACTIONS *
 338              **************************/
 339
 340             /* Compute parameters for interactions between i and j atoms */
 341             qq10             = _mm_mul_pd(iq1,jq0);
 342
 343             /* COULOMB ELECTROSTATICS */
 344             velec            = _mm_mul_pd(qq10,rinv10);
 345             felec            = _mm_mul_pd(velec,rinvsq10);
 346
 347             /* Update potential sum for this i atom from the interaction with this j atom. */
 348             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 349             velecsum         = _mm_add_pd(velecsum,velec);
 350
 351             fscal            = felec;
 352
 353             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 354
 355             /* Calculate temporary vectorial force */
 356             tx               = _mm_mul_pd(fscal,dx10);
 357             ty               = _mm_mul_pd(fscal,dy10);
 358             tz               = _mm_mul_pd(fscal,dz10);
 359
 360             /* Update vectorial force */
 361             fix1             = _mm_add_pd(fix1,tx);
 362             fiy1             = _mm_add_pd(fiy1,ty);
 363             fiz1             = _mm_add_pd(fiz1,tz);
 364
 365             fjx0             = _mm_add_pd(fjx0,tx);
 366             fjy0             = _mm_add_pd(fjy0,ty);
 367             fjz0             = _mm_add_pd(fjz0,tz);
 368
 369             /**************************
 370              * CALCULATE INTERACTIONS *
 371              **************************/
 372
 373             /* Compute parameters for interactions between i and j atoms */
 374             qq20             = _mm_mul_pd(iq2,jq0);
 375
 376             /* COULOMB ELECTROSTATICS */
 377             velec            = _mm_mul_pd(qq20,rinv20);
 378             felec            = _mm_mul_pd(velec,rinvsq20);
 379
 380             /* Update potential sum for this i atom from the interaction with this j atom. */
 381             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 382             velecsum         = _mm_add_pd(velecsum,velec);
 383
 384             fscal            = felec;
 385
 386             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 387
 388             /* Calculate temporary vectorial force */
 389             tx               = _mm_mul_pd(fscal,dx20);
 390             ty               = _mm_mul_pd(fscal,dy20);
 391             tz               = _mm_mul_pd(fscal,dz20);
 392
 393             /* Update vectorial force */
 394             fix2             = _mm_add_pd(fix2,tx);
 395             fiy2             = _mm_add_pd(fiy2,ty);
 396             fiz2             = _mm_add_pd(fiz2,tz);
 397
 398             fjx0             = _mm_add_pd(fjx0,tx);
 399             fjy0             = _mm_add_pd(fjy0,ty);
 400             fjz0             = _mm_add_pd(fjz0,tz);
 401
 402             /**************************
 403              * CALCULATE INTERACTIONS *
 404              **************************/
 405
 406             /* Compute parameters for interactions between i and j atoms */
 407             qq30             = _mm_mul_pd(iq3,jq0);
 408
 409             /* COULOMB ELECTROSTATICS */
 410             velec            = _mm_mul_pd(qq30,rinv30);
 411             felec            = _mm_mul_pd(velec,rinvsq30);
 412
 413             /* Update potential sum for this i atom from the interaction with this j atom. */
 414             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 415             velecsum         = _mm_add_pd(velecsum,velec);
 416
 417             fscal            = felec;
 418
 419             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 420
 421             /* Calculate temporary vectorial force */
 422             tx               = _mm_mul_pd(fscal,dx30);
 423             ty               = _mm_mul_pd(fscal,dy30);
 424             tz               = _mm_mul_pd(fscal,dz30);
 425
 426             /* Update vectorial force */
 427             fix3             = _mm_add_pd(fix3,tx);
 428             fiy3             = _mm_add_pd(fiy3,ty);
 429             fiz3             = _mm_add_pd(fiz3,tz);
 430
 431             fjx0             = _mm_add_pd(fjx0,tx);
 432             fjy0             = _mm_add_pd(fjy0,ty);
 433             fjz0             = _mm_add_pd(fjz0,tz);
 434
 435             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 436
 437             /* Inner loop uses 87 flops */
 438         }
 439
 440         /* End of innermost loop */
 441
 442         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 443                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 444
 445         ggid                        = gid[iidx];
 446         /* Update potential energies */
 447         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 448
 449         /* Increment number of inner iterations */
 450         inneriter                  += j_index_end - j_index_start;
 451
 452         /* Outer loop uses 19 flops */
 453     }
 454
 455     /* Increment number of outer iterations */
 456     outeriter        += nri;
 457
 458     /* Update outer/inner flops */
 459
 460     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*87);
 461 }
 462 /*
 463  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomW4P1_F_sse4_1_double
 464  * Electrostatics interaction: Coulomb
 465  * VdW interaction:            None
 466  * Geometry:                   Water4-Particle
 467  * Calculate force/pot:        Force
 468  */
 469 void
 470 nb_kernel_ElecCoul_VdwNone_GeomW4P1_F_sse4_1_double
 471                     (t_nblist                    * gmx_restrict       nlist,
 472                      rvec                        * gmx_restrict          xx,
 473                      rvec                        * gmx_restrict          ff,
 474                      struct t_forcerec           * gmx_restrict          fr,
 475                      t_mdatoms                   * gmx_restrict     mdatoms,
 476                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 477                      t_nrnb                      * gmx_restrict        nrnb)
 478 {
 479     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 480      * just 0 for non-waters.
 481      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 482      * jnr indices corresponding to data put in the four positions in the SIMD register.
 483      */
 484     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 485     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 486     int              jnrA,jnrB;
 487     int              j_coord_offsetA,j_coord_offsetB;
 488     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 489     real             rcutoff_scalar;
 490     real             *shiftvec,*fshift,*x,*f;
 491     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 492     int              vdwioffset1;
 493     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 494     int              vdwioffset2;
 495     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 496     int              vdwioffset3;
 497     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 498     int              vdwjidx0A,vdwjidx0B;
 499     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 500     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 501     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 502     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 503     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 504     real             *charge;
 505     __m128d          dummy_mask,cutoff_mask;
 506     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 507     __m128d          one     = _mm_set1_pd(1.0);
 508     __m128d          two     = _mm_set1_pd(2.0);
 509     x                = xx[0];
 510     f                = ff[0];
 511
 512     nri              = nlist->nri;
 513     iinr             = nlist->iinr;
 514     jindex           = nlist->jindex;
 515     jjnr             = nlist->jjnr;
 516     shiftidx         = nlist->shift;
 517     gid              = nlist->gid;
 518     shiftvec         = fr->shift_vec[0];
 519     fshift           = fr->fshift[0];
 520     facel            = _mm_set1_pd(fr->ic->epsfac);
 521     charge           = mdatoms->chargeA;
 522
 523     /* Setup water-specific parameters */
 524     inr              = nlist->iinr[0];
 525     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 526     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 527     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 528
 529     /* Avoid stupid compiler warnings */
 530     jnrA = jnrB = 0;
 531     j_coord_offsetA = 0;
 532     j_coord_offsetB = 0;
 533
 534     outeriter        = 0;
 535     inneriter        = 0;
 536
 537     /* Start outer loop over neighborlists */
 538     for(iidx=0; iidx<nri; iidx++)
 539     {
 540         /* Load shift vector for this list */
 541         i_shift_offset   = DIM*shiftidx[iidx];
 542
 543         /* Load limits for loop over neighbors */
 544         j_index_start    = jindex[iidx];
 545         j_index_end      = jindex[iidx+1];
 546
 547         /* Get outer coordinate index */
 548         inr              = iinr[iidx];
 549         i_coord_offset   = DIM*inr;
 550
 551         /* Load i particle coords and add shift vector */
 552         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 553                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 554
 555         fix1             = _mm_setzero_pd();
 556         fiy1             = _mm_setzero_pd();
 557         fiz1             = _mm_setzero_pd();
 558         fix2             = _mm_setzero_pd();
 559         fiy2             = _mm_setzero_pd();
 560         fiz2             = _mm_setzero_pd();
 561         fix3             = _mm_setzero_pd();
 562         fiy3             = _mm_setzero_pd();
 563         fiz3             = _mm_setzero_pd();
 564
 565         /* Start inner kernel loop */
 566         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 567         {
 568
 569             /* Get j neighbor index, and coordinate index */
 570             jnrA             = jjnr[jidx];
 571             jnrB             = jjnr[jidx+1];
 572             j_coord_offsetA  = DIM*jnrA;
 573             j_coord_offsetB  = DIM*jnrB;
 574
 575             /* load j atom coordinates */
 576             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 577                                               &jx0,&jy0,&jz0);
 578
 579             /* Calculate displacement vector */
 580             dx10             = _mm_sub_pd(ix1,jx0);
 581             dy10             = _mm_sub_pd(iy1,jy0);
 582             dz10             = _mm_sub_pd(iz1,jz0);
 583             dx20             = _mm_sub_pd(ix2,jx0);
 584             dy20             = _mm_sub_pd(iy2,jy0);
 585             dz20             = _mm_sub_pd(iz2,jz0);
 586             dx30             = _mm_sub_pd(ix3,jx0);
 587             dy30             = _mm_sub_pd(iy3,jy0);
 588             dz30             = _mm_sub_pd(iz3,jz0);
 589
 590             /* Calculate squared distance and things based on it */
 591             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 592             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 593             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 594
 595             rinv10           = sse41_invsqrt_d(rsq10);
 596             rinv20           = sse41_invsqrt_d(rsq20);
 597             rinv30           = sse41_invsqrt_d(rsq30);
 598
 599             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 600             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 601             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 602
 603             /* Load parameters for j particles */
 604             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 605
 606             fjx0             = _mm_setzero_pd();
 607             fjy0             = _mm_setzero_pd();
 608             fjz0             = _mm_setzero_pd();
 609
 610             /**************************
 611              * CALCULATE INTERACTIONS *
 612              **************************/
 613
 614             /* Compute parameters for interactions between i and j atoms */
 615             qq10             = _mm_mul_pd(iq1,jq0);
 616
 617             /* COULOMB ELECTROSTATICS */
 618             velec            = _mm_mul_pd(qq10,rinv10);
 619             felec            = _mm_mul_pd(velec,rinvsq10);
 620
 621             fscal            = felec;
 622
 623             /* Calculate temporary vectorial force */
 624             tx               = _mm_mul_pd(fscal,dx10);
 625             ty               = _mm_mul_pd(fscal,dy10);
 626             tz               = _mm_mul_pd(fscal,dz10);
 627
 628             /* Update vectorial force */
 629             fix1             = _mm_add_pd(fix1,tx);
 630             fiy1             = _mm_add_pd(fiy1,ty);
 631             fiz1             = _mm_add_pd(fiz1,tz);
 632
 633             fjx0             = _mm_add_pd(fjx0,tx);
 634             fjy0             = _mm_add_pd(fjy0,ty);
 635             fjz0             = _mm_add_pd(fjz0,tz);
 636
 637             /**************************
 638              * CALCULATE INTERACTIONS *
 639              **************************/
 640
 641             /* Compute parameters for interactions between i and j atoms */
 642             qq20             = _mm_mul_pd(iq2,jq0);
 643
 644             /* COULOMB ELECTROSTATICS */
 645             velec            = _mm_mul_pd(qq20,rinv20);
 646             felec            = _mm_mul_pd(velec,rinvsq20);
 647
 648             fscal            = felec;
 649
 650             /* Calculate temporary vectorial force */
 651             tx               = _mm_mul_pd(fscal,dx20);
 652             ty               = _mm_mul_pd(fscal,dy20);
 653             tz               = _mm_mul_pd(fscal,dz20);
 654
 655             /* Update vectorial force */
 656             fix2             = _mm_add_pd(fix2,tx);
 657             fiy2             = _mm_add_pd(fiy2,ty);
 658             fiz2             = _mm_add_pd(fiz2,tz);
 659
 660             fjx0             = _mm_add_pd(fjx0,tx);
 661             fjy0             = _mm_add_pd(fjy0,ty);
 662             fjz0             = _mm_add_pd(fjz0,tz);
 663
 664             /**************************
 665              * CALCULATE INTERACTIONS *
 666              **************************/
 667
 668             /* Compute parameters for interactions between i and j atoms */
 669             qq30             = _mm_mul_pd(iq3,jq0);
 670
 671             /* COULOMB ELECTROSTATICS */
 672             velec            = _mm_mul_pd(qq30,rinv30);
 673             felec            = _mm_mul_pd(velec,rinvsq30);
 674
 675             fscal            = felec;
 676
 677             /* Calculate temporary vectorial force */
 678             tx               = _mm_mul_pd(fscal,dx30);
 679             ty               = _mm_mul_pd(fscal,dy30);
 680             tz               = _mm_mul_pd(fscal,dz30);
 681
 682             /* Update vectorial force */
 683             fix3             = _mm_add_pd(fix3,tx);
 684             fiy3             = _mm_add_pd(fiy3,ty);
 685             fiz3             = _mm_add_pd(fiz3,tz);
 686
 687             fjx0             = _mm_add_pd(fjx0,tx);
 688             fjy0             = _mm_add_pd(fjy0,ty);
 689             fjz0             = _mm_add_pd(fjz0,tz);
 690
 691             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 692
 693             /* Inner loop uses 84 flops */
 694         }
 695
 696         if(jidx<j_index_end)
 697         {
 698
 699             jnrA             = jjnr[jidx];
 700             j_coord_offsetA  = DIM*jnrA;
 701
 702             /* load j atom coordinates */
 703             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 704                                               &jx0,&jy0,&jz0);
 705
 706             /* Calculate displacement vector */
 707             dx10             = _mm_sub_pd(ix1,jx0);
 708             dy10             = _mm_sub_pd(iy1,jy0);
 709             dz10             = _mm_sub_pd(iz1,jz0);
 710             dx20             = _mm_sub_pd(ix2,jx0);
 711             dy20             = _mm_sub_pd(iy2,jy0);
 712             dz20             = _mm_sub_pd(iz2,jz0);
 713             dx30             = _mm_sub_pd(ix3,jx0);
 714             dy30             = _mm_sub_pd(iy3,jy0);
 715             dz30             = _mm_sub_pd(iz3,jz0);
 716
 717             /* Calculate squared distance and things based on it */
 718             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 719             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 720             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 721
 722             rinv10           = sse41_invsqrt_d(rsq10);
 723             rinv20           = sse41_invsqrt_d(rsq20);
 724             rinv30           = sse41_invsqrt_d(rsq30);
 725
 726             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 727             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 728             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 729
 730             /* Load parameters for j particles */
 731             jq0              = _mm_load_sd(charge+jnrA+0);
 732
 733             fjx0             = _mm_setzero_pd();
 734             fjy0             = _mm_setzero_pd();
 735             fjz0             = _mm_setzero_pd();
 736
 737             /**************************
 738              * CALCULATE INTERACTIONS *
 739              **************************/
 740
 741             /* Compute parameters for interactions between i and j atoms */
 742             qq10             = _mm_mul_pd(iq1,jq0);
 743
 744             /* COULOMB ELECTROSTATICS */
 745             velec            = _mm_mul_pd(qq10,rinv10);
 746             felec            = _mm_mul_pd(velec,rinvsq10);
 747
 748             fscal            = felec;
 749
 750             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 751
 752             /* Calculate temporary vectorial force */
 753             tx               = _mm_mul_pd(fscal,dx10);
 754             ty               = _mm_mul_pd(fscal,dy10);
 755             tz               = _mm_mul_pd(fscal,dz10);
 756
 757             /* Update vectorial force */
 758             fix1             = _mm_add_pd(fix1,tx);
 759             fiy1             = _mm_add_pd(fiy1,ty);
 760             fiz1             = _mm_add_pd(fiz1,tz);
 761
 762             fjx0             = _mm_add_pd(fjx0,tx);
 763             fjy0             = _mm_add_pd(fjy0,ty);
 764             fjz0             = _mm_add_pd(fjz0,tz);
 765
 766             /**************************
 767              * CALCULATE INTERACTIONS *
 768              **************************/
 769
 770             /* Compute parameters for interactions between i and j atoms */
 771             qq20             = _mm_mul_pd(iq2,jq0);
 772
 773             /* COULOMB ELECTROSTATICS */
 774             velec            = _mm_mul_pd(qq20,rinv20);
 775             felec            = _mm_mul_pd(velec,rinvsq20);
 776
 777             fscal            = felec;
 778
 779             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 780
 781             /* Calculate temporary vectorial force */
 782             tx               = _mm_mul_pd(fscal,dx20);
 783             ty               = _mm_mul_pd(fscal,dy20);
 784             tz               = _mm_mul_pd(fscal,dz20);
 785
 786             /* Update vectorial force */
 787             fix2             = _mm_add_pd(fix2,tx);
 788             fiy2             = _mm_add_pd(fiy2,ty);
 789             fiz2             = _mm_add_pd(fiz2,tz);
 790
 791             fjx0             = _mm_add_pd(fjx0,tx);
 792             fjy0             = _mm_add_pd(fjy0,ty);
 793             fjz0             = _mm_add_pd(fjz0,tz);
 794
 795             /**************************
 796              * CALCULATE INTERACTIONS *
 797              **************************/
 798
 799             /* Compute parameters for interactions between i and j atoms */
 800             qq30             = _mm_mul_pd(iq3,jq0);
 801
 802             /* COULOMB ELECTROSTATICS */
 803             velec            = _mm_mul_pd(qq30,rinv30);
 804             felec            = _mm_mul_pd(velec,rinvsq30);
 805
 806             fscal            = felec;
 807
 808             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 809
 810             /* Calculate temporary vectorial force */
 811             tx               = _mm_mul_pd(fscal,dx30);
 812             ty               = _mm_mul_pd(fscal,dy30);
 813             tz               = _mm_mul_pd(fscal,dz30);
 814
 815             /* Update vectorial force */
 816             fix3             = _mm_add_pd(fix3,tx);
 817             fiy3             = _mm_add_pd(fiy3,ty);
 818             fiz3             = _mm_add_pd(fiz3,tz);
 819
 820             fjx0             = _mm_add_pd(fjx0,tx);
 821             fjy0             = _mm_add_pd(fjy0,ty);
 822             fjz0             = _mm_add_pd(fjz0,tz);
 823
 824             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 825
 826             /* Inner loop uses 84 flops */
 827         }
 828
 829         /* End of innermost loop */
 830
 831         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 832                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 833
 834         /* Increment number of inner iterations */
 835         inneriter                  += j_index_end - j_index_start;
 836
 837         /* Outer loop uses 18 flops */
 838     }
 839
 840     /* Increment number of outer iterations */
 841     outeriter        += nri;
 842
 843     /* Update outer/inner flops */
 844
 845     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*84);
 846 }