src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCoul_VdwNone_GeomW3P1_avx_128_fma_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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 "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_avx_128_fma_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomW3P1_VF_avx_128_fma_double
  51  * Electrostatics interaction: Coulomb
  52  * VdW interaction:            None
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCoul_VdwNone_GeomW3P1_VF_avx_128_fma_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              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwioffset1;
  82     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  83     int              vdwioffset2;
  84     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  85     int              vdwjidx0A,vdwjidx0B;
  86     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  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          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     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 113     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 114     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 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,
 140                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 141
 142         fix0             = _mm_setzero_pd();
 143         fiy0             = _mm_setzero_pd();
 144         fiz0             = _mm_setzero_pd();
 145         fix1             = _mm_setzero_pd();
 146         fiy1             = _mm_setzero_pd();
 147         fiz1             = _mm_setzero_pd();
 148         fix2             = _mm_setzero_pd();
 149         fiy2             = _mm_setzero_pd();
 150         fiz2             = _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             dx00             = _mm_sub_pd(ix0,jx0);
 171             dy00             = _mm_sub_pd(iy0,jy0);
 172             dz00             = _mm_sub_pd(iz0,jz0);
 173             dx10             = _mm_sub_pd(ix1,jx0);
 174             dy10             = _mm_sub_pd(iy1,jy0);
 175             dz10             = _mm_sub_pd(iz1,jz0);
 176             dx20             = _mm_sub_pd(ix2,jx0);
 177             dy20             = _mm_sub_pd(iy2,jy0);
 178             dz20             = _mm_sub_pd(iz2,jz0);
 179
 180             /* Calculate squared distance and things based on it */
 181             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 182             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 183             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 184
 185             rinv00           = avx128fma_invsqrt_d(rsq00);
 186             rinv10           = avx128fma_invsqrt_d(rsq10);
 187             rinv20           = avx128fma_invsqrt_d(rsq20);
 188
 189             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 190             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 191             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 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             qq00             = _mm_mul_pd(iq0,jq0);
 206
 207             /* COULOMB ELECTROSTATICS */
 208             velec            = _mm_mul_pd(qq00,rinv00);
 209             felec            = _mm_mul_pd(velec,rinvsq00);
 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             /* Update vectorial force */
 217             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 218             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 219             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 220
 221             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 222             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 223             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 224
 225             /**************************
 226              * CALCULATE INTERACTIONS *
 227              **************************/
 228
 229             /* Compute parameters for interactions between i and j atoms */
 230             qq10             = _mm_mul_pd(iq1,jq0);
 231
 232             /* COULOMB ELECTROSTATICS */
 233             velec            = _mm_mul_pd(qq10,rinv10);
 234             felec            = _mm_mul_pd(velec,rinvsq10);
 235
 236             /* Update potential sum for this i atom from the interaction with this j atom. */
 237             velecsum         = _mm_add_pd(velecsum,velec);
 238
 239             fscal            = felec;
 240
 241             /* Update vectorial force */
 242             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 243             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 244             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 245
 246             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 247             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 248             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 249
 250             /**************************
 251              * CALCULATE INTERACTIONS *
 252              **************************/
 253
 254             /* Compute parameters for interactions between i and j atoms */
 255             qq20             = _mm_mul_pd(iq2,jq0);
 256
 257             /* COULOMB ELECTROSTATICS */
 258             velec            = _mm_mul_pd(qq20,rinv20);
 259             felec            = _mm_mul_pd(velec,rinvsq20);
 260
 261             /* Update potential sum for this i atom from the interaction with this j atom. */
 262             velecsum         = _mm_add_pd(velecsum,velec);
 263
 264             fscal            = felec;
 265
 266             /* Update vectorial force */
 267             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 268             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 269             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 270
 271             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 272             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 273             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 274
 275             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 276
 277             /* Inner loop uses 96 flops */
 278         }
 279
 280         if(jidx<j_index_end)
 281         {
 282
 283             jnrA             = jjnr[jidx];
 284             j_coord_offsetA  = DIM*jnrA;
 285
 286             /* load j atom coordinates */
 287             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 288                                               &jx0,&jy0,&jz0);
 289
 290             /* Calculate displacement vector */
 291             dx00             = _mm_sub_pd(ix0,jx0);
 292             dy00             = _mm_sub_pd(iy0,jy0);
 293             dz00             = _mm_sub_pd(iz0,jz0);
 294             dx10             = _mm_sub_pd(ix1,jx0);
 295             dy10             = _mm_sub_pd(iy1,jy0);
 296             dz10             = _mm_sub_pd(iz1,jz0);
 297             dx20             = _mm_sub_pd(ix2,jx0);
 298             dy20             = _mm_sub_pd(iy2,jy0);
 299             dz20             = _mm_sub_pd(iz2,jz0);
 300
 301             /* Calculate squared distance and things based on it */
 302             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 303             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 304             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 305
 306             rinv00           = avx128fma_invsqrt_d(rsq00);
 307             rinv10           = avx128fma_invsqrt_d(rsq10);
 308             rinv20           = avx128fma_invsqrt_d(rsq20);
 309
 310             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 311             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 312             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 313
 314             /* Load parameters for j particles */
 315             jq0              = _mm_load_sd(charge+jnrA+0);
 316
 317             fjx0             = _mm_setzero_pd();
 318             fjy0             = _mm_setzero_pd();
 319             fjz0             = _mm_setzero_pd();
 320
 321             /**************************
 322              * CALCULATE INTERACTIONS *
 323              **************************/
 324
 325             /* Compute parameters for interactions between i and j atoms */
 326             qq00             = _mm_mul_pd(iq0,jq0);
 327
 328             /* COULOMB ELECTROSTATICS */
 329             velec            = _mm_mul_pd(qq00,rinv00);
 330             felec            = _mm_mul_pd(velec,rinvsq00);
 331
 332             /* Update potential sum for this i atom from the interaction with this j atom. */
 333             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 334             velecsum         = _mm_add_pd(velecsum,velec);
 335
 336             fscal            = felec;
 337
 338             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 339
 340             /* Update vectorial force */
 341             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 342             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 343             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 344
 345             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 346             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 347             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 348
 349             /**************************
 350              * CALCULATE INTERACTIONS *
 351              **************************/
 352
 353             /* Compute parameters for interactions between i and j atoms */
 354             qq10             = _mm_mul_pd(iq1,jq0);
 355
 356             /* COULOMB ELECTROSTATICS */
 357             velec            = _mm_mul_pd(qq10,rinv10);
 358             felec            = _mm_mul_pd(velec,rinvsq10);
 359
 360             /* Update potential sum for this i atom from the interaction with this j atom. */
 361             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 362             velecsum         = _mm_add_pd(velecsum,velec);
 363
 364             fscal            = felec;
 365
 366             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 367
 368             /* Update vectorial force */
 369             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 370             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 371             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 372
 373             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 374             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 375             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 376
 377             /**************************
 378              * CALCULATE INTERACTIONS *
 379              **************************/
 380
 381             /* Compute parameters for interactions between i and j atoms */
 382             qq20             = _mm_mul_pd(iq2,jq0);
 383
 384             /* COULOMB ELECTROSTATICS */
 385             velec            = _mm_mul_pd(qq20,rinv20);
 386             felec            = _mm_mul_pd(velec,rinvsq20);
 387
 388             /* Update potential sum for this i atom from the interaction with this j atom. */
 389             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 390             velecsum         = _mm_add_pd(velecsum,velec);
 391
 392             fscal            = felec;
 393
 394             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 395
 396             /* Update vectorial force */
 397             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 398             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 399             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 400
 401             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 402             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 403             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 404
 405             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 406
 407             /* Inner loop uses 96 flops */
 408         }
 409
 410         /* End of innermost loop */
 411
 412         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 413                                               f+i_coord_offset,fshift+i_shift_offset);
 414
 415         ggid                        = gid[iidx];
 416         /* Update potential energies */
 417         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 418
 419         /* Increment number of inner iterations */
 420         inneriter                  += j_index_end - j_index_start;
 421
 422         /* Outer loop uses 19 flops */
 423     }
 424
 425     /* Increment number of outer iterations */
 426     outeriter        += nri;
 427
 428     /* Update outer/inner flops */
 429
 430     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*96);
 431 }
 432 /*
 433  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomW3P1_F_avx_128_fma_double
 434  * Electrostatics interaction: Coulomb
 435  * VdW interaction:            None
 436  * Geometry:                   Water3-Particle
 437  * Calculate force/pot:        Force
 438  */
 439 void
 440 nb_kernel_ElecCoul_VdwNone_GeomW3P1_F_avx_128_fma_double
 441                     (t_nblist                    * gmx_restrict       nlist,
 442                      rvec                        * gmx_restrict          xx,
 443                      rvec                        * gmx_restrict          ff,
 444                      struct t_forcerec           * gmx_restrict          fr,
 445                      t_mdatoms                   * gmx_restrict     mdatoms,
 446                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 447                      t_nrnb                      * gmx_restrict        nrnb)
 448 {
 449     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 450      * just 0 for non-waters.
 451      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 452      * jnr indices corresponding to data put in the four positions in the SIMD register.
 453      */
 454     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 455     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 456     int              jnrA,jnrB;
 457     int              j_coord_offsetA,j_coord_offsetB;
 458     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 459     real             rcutoff_scalar;
 460     real             *shiftvec,*fshift,*x,*f;
 461     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 462     int              vdwioffset0;
 463     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 464     int              vdwioffset1;
 465     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 466     int              vdwioffset2;
 467     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 468     int              vdwjidx0A,vdwjidx0B;
 469     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 470     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 471     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 472     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 473     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 474     real             *charge;
 475     __m128d          dummy_mask,cutoff_mask;
 476     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 477     __m128d          one     = _mm_set1_pd(1.0);
 478     __m128d          two     = _mm_set1_pd(2.0);
 479     x                = xx[0];
 480     f                = ff[0];
 481
 482     nri              = nlist->nri;
 483     iinr             = nlist->iinr;
 484     jindex           = nlist->jindex;
 485     jjnr             = nlist->jjnr;
 486     shiftidx         = nlist->shift;
 487     gid              = nlist->gid;
 488     shiftvec         = fr->shift_vec[0];
 489     fshift           = fr->fshift[0];
 490     facel            = _mm_set1_pd(fr->ic->epsfac);
 491     charge           = mdatoms->chargeA;
 492
 493     /* Setup water-specific parameters */
 494     inr              = nlist->iinr[0];
 495     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 496     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 497     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 498
 499     /* Avoid stupid compiler warnings */
 500     jnrA = jnrB = 0;
 501     j_coord_offsetA = 0;
 502     j_coord_offsetB = 0;
 503
 504     outeriter        = 0;
 505     inneriter        = 0;
 506
 507     /* Start outer loop over neighborlists */
 508     for(iidx=0; iidx<nri; iidx++)
 509     {
 510         /* Load shift vector for this list */
 511         i_shift_offset   = DIM*shiftidx[iidx];
 512
 513         /* Load limits for loop over neighbors */
 514         j_index_start    = jindex[iidx];
 515         j_index_end      = jindex[iidx+1];
 516
 517         /* Get outer coordinate index */
 518         inr              = iinr[iidx];
 519         i_coord_offset   = DIM*inr;
 520
 521         /* Load i particle coords and add shift vector */
 522         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 523                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 524
 525         fix0             = _mm_setzero_pd();
 526         fiy0             = _mm_setzero_pd();
 527         fiz0             = _mm_setzero_pd();
 528         fix1             = _mm_setzero_pd();
 529         fiy1             = _mm_setzero_pd();
 530         fiz1             = _mm_setzero_pd();
 531         fix2             = _mm_setzero_pd();
 532         fiy2             = _mm_setzero_pd();
 533         fiz2             = _mm_setzero_pd();
 534
 535         /* Start inner kernel loop */
 536         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 537         {
 538
 539             /* Get j neighbor index, and coordinate index */
 540             jnrA             = jjnr[jidx];
 541             jnrB             = jjnr[jidx+1];
 542             j_coord_offsetA  = DIM*jnrA;
 543             j_coord_offsetB  = DIM*jnrB;
 544
 545             /* load j atom coordinates */
 546             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 547                                               &jx0,&jy0,&jz0);
 548
 549             /* Calculate displacement vector */
 550             dx00             = _mm_sub_pd(ix0,jx0);
 551             dy00             = _mm_sub_pd(iy0,jy0);
 552             dz00             = _mm_sub_pd(iz0,jz0);
 553             dx10             = _mm_sub_pd(ix1,jx0);
 554             dy10             = _mm_sub_pd(iy1,jy0);
 555             dz10             = _mm_sub_pd(iz1,jz0);
 556             dx20             = _mm_sub_pd(ix2,jx0);
 557             dy20             = _mm_sub_pd(iy2,jy0);
 558             dz20             = _mm_sub_pd(iz2,jz0);
 559
 560             /* Calculate squared distance and things based on it */
 561             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 562             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 563             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 564
 565             rinv00           = avx128fma_invsqrt_d(rsq00);
 566             rinv10           = avx128fma_invsqrt_d(rsq10);
 567             rinv20           = avx128fma_invsqrt_d(rsq20);
 568
 569             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 570             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 571             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 572
 573             /* Load parameters for j particles */
 574             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 575
 576             fjx0             = _mm_setzero_pd();
 577             fjy0             = _mm_setzero_pd();
 578             fjz0             = _mm_setzero_pd();
 579
 580             /**************************
 581              * CALCULATE INTERACTIONS *
 582              **************************/
 583
 584             /* Compute parameters for interactions between i and j atoms */
 585             qq00             = _mm_mul_pd(iq0,jq0);
 586
 587             /* COULOMB ELECTROSTATICS */
 588             velec            = _mm_mul_pd(qq00,rinv00);
 589             felec            = _mm_mul_pd(velec,rinvsq00);
 590
 591             fscal            = felec;
 592
 593             /* Update vectorial force */
 594             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 595             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 596             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 597
 598             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 599             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 600             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 601
 602             /**************************
 603              * CALCULATE INTERACTIONS *
 604              **************************/
 605
 606             /* Compute parameters for interactions between i and j atoms */
 607             qq10             = _mm_mul_pd(iq1,jq0);
 608
 609             /* COULOMB ELECTROSTATICS */
 610             velec            = _mm_mul_pd(qq10,rinv10);
 611             felec            = _mm_mul_pd(velec,rinvsq10);
 612
 613             fscal            = felec;
 614
 615             /* Update vectorial force */
 616             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 617             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 618             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 619
 620             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 621             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 622             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 623
 624             /**************************
 625              * CALCULATE INTERACTIONS *
 626              **************************/
 627
 628             /* Compute parameters for interactions between i and j atoms */
 629             qq20             = _mm_mul_pd(iq2,jq0);
 630
 631             /* COULOMB ELECTROSTATICS */
 632             velec            = _mm_mul_pd(qq20,rinv20);
 633             felec            = _mm_mul_pd(velec,rinvsq20);
 634
 635             fscal            = felec;
 636
 637             /* Update vectorial force */
 638             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 639             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 640             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 641
 642             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 643             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 644             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 645
 646             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 647
 648             /* Inner loop uses 93 flops */
 649         }
 650
 651         if(jidx<j_index_end)
 652         {
 653
 654             jnrA             = jjnr[jidx];
 655             j_coord_offsetA  = DIM*jnrA;
 656
 657             /* load j atom coordinates */
 658             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 659                                               &jx0,&jy0,&jz0);
 660
 661             /* Calculate displacement vector */
 662             dx00             = _mm_sub_pd(ix0,jx0);
 663             dy00             = _mm_sub_pd(iy0,jy0);
 664             dz00             = _mm_sub_pd(iz0,jz0);
 665             dx10             = _mm_sub_pd(ix1,jx0);
 666             dy10             = _mm_sub_pd(iy1,jy0);
 667             dz10             = _mm_sub_pd(iz1,jz0);
 668             dx20             = _mm_sub_pd(ix2,jx0);
 669             dy20             = _mm_sub_pd(iy2,jy0);
 670             dz20             = _mm_sub_pd(iz2,jz0);
 671
 672             /* Calculate squared distance and things based on it */
 673             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 674             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 675             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 676
 677             rinv00           = avx128fma_invsqrt_d(rsq00);
 678             rinv10           = avx128fma_invsqrt_d(rsq10);
 679             rinv20           = avx128fma_invsqrt_d(rsq20);
 680
 681             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 682             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 683             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 684
 685             /* Load parameters for j particles */
 686             jq0              = _mm_load_sd(charge+jnrA+0);
 687
 688             fjx0             = _mm_setzero_pd();
 689             fjy0             = _mm_setzero_pd();
 690             fjz0             = _mm_setzero_pd();
 691
 692             /**************************
 693              * CALCULATE INTERACTIONS *
 694              **************************/
 695
 696             /* Compute parameters for interactions between i and j atoms */
 697             qq00             = _mm_mul_pd(iq0,jq0);
 698
 699             /* COULOMB ELECTROSTATICS */
 700             velec            = _mm_mul_pd(qq00,rinv00);
 701             felec            = _mm_mul_pd(velec,rinvsq00);
 702
 703             fscal            = felec;
 704
 705             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 706
 707             /* Update vectorial force */
 708             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 709             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 710             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 711
 712             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 713             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 714             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 715
 716             /**************************
 717              * CALCULATE INTERACTIONS *
 718              **************************/
 719
 720             /* Compute parameters for interactions between i and j atoms */
 721             qq10             = _mm_mul_pd(iq1,jq0);
 722
 723             /* COULOMB ELECTROSTATICS */
 724             velec            = _mm_mul_pd(qq10,rinv10);
 725             felec            = _mm_mul_pd(velec,rinvsq10);
 726
 727             fscal            = felec;
 728
 729             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 730
 731             /* Update vectorial force */
 732             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 733             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 734             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 735
 736             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 737             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 738             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 739
 740             /**************************
 741              * CALCULATE INTERACTIONS *
 742              **************************/
 743
 744             /* Compute parameters for interactions between i and j atoms */
 745             qq20             = _mm_mul_pd(iq2,jq0);
 746
 747             /* COULOMB ELECTROSTATICS */
 748             velec            = _mm_mul_pd(qq20,rinv20);
 749             felec            = _mm_mul_pd(velec,rinvsq20);
 750
 751             fscal            = felec;
 752
 753             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 754
 755             /* Update vectorial force */
 756             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 757             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 758             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 759
 760             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 761             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 762             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 763
 764             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 765
 766             /* Inner loop uses 93 flops */
 767         }
 768
 769         /* End of innermost loop */
 770
 771         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 772                                               f+i_coord_offset,fshift+i_shift_offset);
 773
 774         /* Increment number of inner iterations */
 775         inneriter                  += j_index_end - j_index_start;
 776
 777         /* Outer loop uses 18 flops */
 778     }
 779
 780     /* Increment number of outer iterations */
 781     outeriter        += nri;
 782
 783     /* Update outer/inner flops */
 784
 785     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*93);
 786 }