src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecRF_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_ElecRF_VdwNone_GeomW3P1_VF_avx_128_fma_double
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            None
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRF_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     krf              = _mm_set1_pd(fr->ic->k_rf);
 110     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 111     crf              = _mm_set1_pd(fr->ic->c_rf);
 112
 113     /* Setup water-specific parameters */
 114     inr              = nlist->iinr[0];
 115     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+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
 119     /* Avoid stupid compiler warnings */
 120     jnrA = jnrB = 0;
 121     j_coord_offsetA = 0;
 122     j_coord_offsetB = 0;
 123
 124     outeriter        = 0;
 125     inneriter        = 0;
 126
 127     /* Start outer loop over neighborlists */
 128     for(iidx=0; iidx<nri; iidx++)
 129     {
 130         /* Load shift vector for this list */
 131         i_shift_offset   = DIM*shiftidx[iidx];
 132
 133         /* Load limits for loop over neighbors */
 134         j_index_start    = jindex[iidx];
 135         j_index_end      = jindex[iidx+1];
 136
 137         /* Get outer coordinate index */
 138         inr              = iinr[iidx];
 139         i_coord_offset   = DIM*inr;
 140
 141         /* Load i particle coords and add shift vector */
 142         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 143                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 144
 145         fix0             = _mm_setzero_pd();
 146         fiy0             = _mm_setzero_pd();
 147         fiz0             = _mm_setzero_pd();
 148         fix1             = _mm_setzero_pd();
 149         fiy1             = _mm_setzero_pd();
 150         fiz1             = _mm_setzero_pd();
 151         fix2             = _mm_setzero_pd();
 152         fiy2             = _mm_setzero_pd();
 153         fiz2             = _mm_setzero_pd();
 154
 155         /* Reset potential sums */
 156         velecsum         = _mm_setzero_pd();
 157
 158         /* Start inner kernel loop */
 159         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 160         {
 161
 162             /* Get j neighbor index, and coordinate index */
 163             jnrA             = jjnr[jidx];
 164             jnrB             = jjnr[jidx+1];
 165             j_coord_offsetA  = DIM*jnrA;
 166             j_coord_offsetB  = DIM*jnrB;
 167
 168             /* load j atom coordinates */
 169             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 170                                               &jx0,&jy0,&jz0);
 171
 172             /* Calculate displacement vector */
 173             dx00             = _mm_sub_pd(ix0,jx0);
 174             dy00             = _mm_sub_pd(iy0,jy0);
 175             dz00             = _mm_sub_pd(iz0,jz0);
 176             dx10             = _mm_sub_pd(ix1,jx0);
 177             dy10             = _mm_sub_pd(iy1,jy0);
 178             dz10             = _mm_sub_pd(iz1,jz0);
 179             dx20             = _mm_sub_pd(ix2,jx0);
 180             dy20             = _mm_sub_pd(iy2,jy0);
 181             dz20             = _mm_sub_pd(iz2,jz0);
 182
 183             /* Calculate squared distance and things based on it */
 184             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 185             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 186             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 187
 188             rinv00           = avx128fma_invsqrt_d(rsq00);
 189             rinv10           = avx128fma_invsqrt_d(rsq10);
 190             rinv20           = avx128fma_invsqrt_d(rsq20);
 191
 192             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 193             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 194             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 195
 196             /* Load parameters for j particles */
 197             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 198
 199             fjx0             = _mm_setzero_pd();
 200             fjy0             = _mm_setzero_pd();
 201             fjz0             = _mm_setzero_pd();
 202
 203             /**************************
 204              * CALCULATE INTERACTIONS *
 205              **************************/
 206
 207             /* Compute parameters for interactions between i and j atoms */
 208             qq00             = _mm_mul_pd(iq0,jq0);
 209
 210             /* REACTION-FIELD ELECTROSTATICS */
 211             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 212             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 213
 214             /* Update potential sum for this i atom from the interaction with this j atom. */
 215             velecsum         = _mm_add_pd(velecsum,velec);
 216
 217             fscal            = felec;
 218
 219             /* Update vectorial force */
 220             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 221             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 222             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 223
 224             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 225             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 226             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 227
 228             /**************************
 229              * CALCULATE INTERACTIONS *
 230              **************************/
 231
 232             /* Compute parameters for interactions between i and j atoms */
 233             qq10             = _mm_mul_pd(iq1,jq0);
 234
 235             /* REACTION-FIELD ELECTROSTATICS */
 236             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 237             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 238
 239             /* Update potential sum for this i atom from the interaction with this j atom. */
 240             velecsum         = _mm_add_pd(velecsum,velec);
 241
 242             fscal            = felec;
 243
 244             /* Update vectorial force */
 245             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 246             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 247             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 248
 249             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 250             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 251             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 252
 253             /**************************
 254              * CALCULATE INTERACTIONS *
 255              **************************/
 256
 257             /* Compute parameters for interactions between i and j atoms */
 258             qq20             = _mm_mul_pd(iq2,jq0);
 259
 260             /* REACTION-FIELD ELECTROSTATICS */
 261             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 262             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 263
 264             /* Update potential sum for this i atom from the interaction with this j atom. */
 265             velecsum         = _mm_add_pd(velecsum,velec);
 266
 267             fscal            = felec;
 268
 269             /* Update vectorial force */
 270             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 271             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 272             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 273
 274             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 275             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 276             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 277
 278             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 279
 280             /* Inner loop uses 108 flops */
 281         }
 282
 283         if(jidx<j_index_end)
 284         {
 285
 286             jnrA             = jjnr[jidx];
 287             j_coord_offsetA  = DIM*jnrA;
 288
 289             /* load j atom coordinates */
 290             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 291                                               &jx0,&jy0,&jz0);
 292
 293             /* Calculate displacement vector */
 294             dx00             = _mm_sub_pd(ix0,jx0);
 295             dy00             = _mm_sub_pd(iy0,jy0);
 296             dz00             = _mm_sub_pd(iz0,jz0);
 297             dx10             = _mm_sub_pd(ix1,jx0);
 298             dy10             = _mm_sub_pd(iy1,jy0);
 299             dz10             = _mm_sub_pd(iz1,jz0);
 300             dx20             = _mm_sub_pd(ix2,jx0);
 301             dy20             = _mm_sub_pd(iy2,jy0);
 302             dz20             = _mm_sub_pd(iz2,jz0);
 303
 304             /* Calculate squared distance and things based on it */
 305             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 306             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 307             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 308
 309             rinv00           = avx128fma_invsqrt_d(rsq00);
 310             rinv10           = avx128fma_invsqrt_d(rsq10);
 311             rinv20           = avx128fma_invsqrt_d(rsq20);
 312
 313             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 314             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 315             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 316
 317             /* Load parameters for j particles */
 318             jq0              = _mm_load_sd(charge+jnrA+0);
 319
 320             fjx0             = _mm_setzero_pd();
 321             fjy0             = _mm_setzero_pd();
 322             fjz0             = _mm_setzero_pd();
 323
 324             /**************************
 325              * CALCULATE INTERACTIONS *
 326              **************************/
 327
 328             /* Compute parameters for interactions between i and j atoms */
 329             qq00             = _mm_mul_pd(iq0,jq0);
 330
 331             /* REACTION-FIELD ELECTROSTATICS */
 332             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 333             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 334
 335             /* Update potential sum for this i atom from the interaction with this j atom. */
 336             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 337             velecsum         = _mm_add_pd(velecsum,velec);
 338
 339             fscal            = felec;
 340
 341             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 342
 343             /* Update vectorial force */
 344             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 345             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 346             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 347
 348             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 349             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 350             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 351
 352             /**************************
 353              * CALCULATE INTERACTIONS *
 354              **************************/
 355
 356             /* Compute parameters for interactions between i and j atoms */
 357             qq10             = _mm_mul_pd(iq1,jq0);
 358
 359             /* REACTION-FIELD ELECTROSTATICS */
 360             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 361             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 362
 363             /* Update potential sum for this i atom from the interaction with this j atom. */
 364             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 365             velecsum         = _mm_add_pd(velecsum,velec);
 366
 367             fscal            = felec;
 368
 369             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 370
 371             /* Update vectorial force */
 372             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 373             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 374             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 375
 376             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 377             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 378             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 379
 380             /**************************
 381              * CALCULATE INTERACTIONS *
 382              **************************/
 383
 384             /* Compute parameters for interactions between i and j atoms */
 385             qq20             = _mm_mul_pd(iq2,jq0);
 386
 387             /* REACTION-FIELD ELECTROSTATICS */
 388             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 389             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 390
 391             /* Update potential sum for this i atom from the interaction with this j atom. */
 392             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 393             velecsum         = _mm_add_pd(velecsum,velec);
 394
 395             fscal            = felec;
 396
 397             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 398
 399             /* Update vectorial force */
 400             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 401             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 402             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 403
 404             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 405             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 406             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 407
 408             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 409
 410             /* Inner loop uses 108 flops */
 411         }
 412
 413         /* End of innermost loop */
 414
 415         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 416                                               f+i_coord_offset,fshift+i_shift_offset);
 417
 418         ggid                        = gid[iidx];
 419         /* Update potential energies */
 420         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 421
 422         /* Increment number of inner iterations */
 423         inneriter                  += j_index_end - j_index_start;
 424
 425         /* Outer loop uses 19 flops */
 426     }
 427
 428     /* Increment number of outer iterations */
 429     outeriter        += nri;
 430
 431     /* Update outer/inner flops */
 432
 433     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*108);
 434 }
 435 /*
 436  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomW3P1_F_avx_128_fma_double
 437  * Electrostatics interaction: ReactionField
 438  * VdW interaction:            None
 439  * Geometry:                   Water3-Particle
 440  * Calculate force/pot:        Force
 441  */
 442 void
 443 nb_kernel_ElecRF_VdwNone_GeomW3P1_F_avx_128_fma_double
 444                     (t_nblist                    * gmx_restrict       nlist,
 445                      rvec                        * gmx_restrict          xx,
 446                      rvec                        * gmx_restrict          ff,
 447                      struct t_forcerec           * gmx_restrict          fr,
 448                      t_mdatoms                   * gmx_restrict     mdatoms,
 449                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 450                      t_nrnb                      * gmx_restrict        nrnb)
 451 {
 452     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 453      * just 0 for non-waters.
 454      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 455      * jnr indices corresponding to data put in the four positions in the SIMD register.
 456      */
 457     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 458     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 459     int              jnrA,jnrB;
 460     int              j_coord_offsetA,j_coord_offsetB;
 461     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 462     real             rcutoff_scalar;
 463     real             *shiftvec,*fshift,*x,*f;
 464     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 465     int              vdwioffset0;
 466     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 467     int              vdwioffset1;
 468     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 469     int              vdwioffset2;
 470     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 471     int              vdwjidx0A,vdwjidx0B;
 472     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 473     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 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          velec,felec,velecsum,facel,crf,krf,krf2;
 477     real             *charge;
 478     __m128d          dummy_mask,cutoff_mask;
 479     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 480     __m128d          one     = _mm_set1_pd(1.0);
 481     __m128d          two     = _mm_set1_pd(2.0);
 482     x                = xx[0];
 483     f                = ff[0];
 484
 485     nri              = nlist->nri;
 486     iinr             = nlist->iinr;
 487     jindex           = nlist->jindex;
 488     jjnr             = nlist->jjnr;
 489     shiftidx         = nlist->shift;
 490     gid              = nlist->gid;
 491     shiftvec         = fr->shift_vec[0];
 492     fshift           = fr->fshift[0];
 493     facel            = _mm_set1_pd(fr->ic->epsfac);
 494     charge           = mdatoms->chargeA;
 495     krf              = _mm_set1_pd(fr->ic->k_rf);
 496     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 497     crf              = _mm_set1_pd(fr->ic->c_rf);
 498
 499     /* Setup water-specific parameters */
 500     inr              = nlist->iinr[0];
 501     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+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
 505     /* Avoid stupid compiler warnings */
 506     jnrA = jnrB = 0;
 507     j_coord_offsetA = 0;
 508     j_coord_offsetB = 0;
 509
 510     outeriter        = 0;
 511     inneriter        = 0;
 512
 513     /* Start outer loop over neighborlists */
 514     for(iidx=0; iidx<nri; iidx++)
 515     {
 516         /* Load shift vector for this list */
 517         i_shift_offset   = DIM*shiftidx[iidx];
 518
 519         /* Load limits for loop over neighbors */
 520         j_index_start    = jindex[iidx];
 521         j_index_end      = jindex[iidx+1];
 522
 523         /* Get outer coordinate index */
 524         inr              = iinr[iidx];
 525         i_coord_offset   = DIM*inr;
 526
 527         /* Load i particle coords and add shift vector */
 528         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 529                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 530
 531         fix0             = _mm_setzero_pd();
 532         fiy0             = _mm_setzero_pd();
 533         fiz0             = _mm_setzero_pd();
 534         fix1             = _mm_setzero_pd();
 535         fiy1             = _mm_setzero_pd();
 536         fiz1             = _mm_setzero_pd();
 537         fix2             = _mm_setzero_pd();
 538         fiy2             = _mm_setzero_pd();
 539         fiz2             = _mm_setzero_pd();
 540
 541         /* Start inner kernel loop */
 542         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 543         {
 544
 545             /* Get j neighbor index, and coordinate index */
 546             jnrA             = jjnr[jidx];
 547             jnrB             = jjnr[jidx+1];
 548             j_coord_offsetA  = DIM*jnrA;
 549             j_coord_offsetB  = DIM*jnrB;
 550
 551             /* load j atom coordinates */
 552             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 553                                               &jx0,&jy0,&jz0);
 554
 555             /* Calculate displacement vector */
 556             dx00             = _mm_sub_pd(ix0,jx0);
 557             dy00             = _mm_sub_pd(iy0,jy0);
 558             dz00             = _mm_sub_pd(iz0,jz0);
 559             dx10             = _mm_sub_pd(ix1,jx0);
 560             dy10             = _mm_sub_pd(iy1,jy0);
 561             dz10             = _mm_sub_pd(iz1,jz0);
 562             dx20             = _mm_sub_pd(ix2,jx0);
 563             dy20             = _mm_sub_pd(iy2,jy0);
 564             dz20             = _mm_sub_pd(iz2,jz0);
 565
 566             /* Calculate squared distance and things based on it */
 567             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 568             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 569             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 570
 571             rinv00           = avx128fma_invsqrt_d(rsq00);
 572             rinv10           = avx128fma_invsqrt_d(rsq10);
 573             rinv20           = avx128fma_invsqrt_d(rsq20);
 574
 575             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 576             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 577             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 578
 579             /* Load parameters for j particles */
 580             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 581
 582             fjx0             = _mm_setzero_pd();
 583             fjy0             = _mm_setzero_pd();
 584             fjz0             = _mm_setzero_pd();
 585
 586             /**************************
 587              * CALCULATE INTERACTIONS *
 588              **************************/
 589
 590             /* Compute parameters for interactions between i and j atoms */
 591             qq00             = _mm_mul_pd(iq0,jq0);
 592
 593             /* REACTION-FIELD ELECTROSTATICS */
 594             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 595
 596             fscal            = felec;
 597
 598             /* Update vectorial force */
 599             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 600             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 601             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 602
 603             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 604             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 605             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 606
 607             /**************************
 608              * CALCULATE INTERACTIONS *
 609              **************************/
 610
 611             /* Compute parameters for interactions between i and j atoms */
 612             qq10             = _mm_mul_pd(iq1,jq0);
 613
 614             /* REACTION-FIELD ELECTROSTATICS */
 615             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 616
 617             fscal            = felec;
 618
 619             /* Update vectorial force */
 620             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 621             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 622             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 623
 624             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 625             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 626             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 627
 628             /**************************
 629              * CALCULATE INTERACTIONS *
 630              **************************/
 631
 632             /* Compute parameters for interactions between i and j atoms */
 633             qq20             = _mm_mul_pd(iq2,jq0);
 634
 635             /* REACTION-FIELD ELECTROSTATICS */
 636             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 637
 638             fscal            = felec;
 639
 640             /* Update vectorial force */
 641             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 642             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 643             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 644
 645             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 646             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 647             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 648
 649             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 650
 651             /* Inner loop uses 93 flops */
 652         }
 653
 654         if(jidx<j_index_end)
 655         {
 656
 657             jnrA             = jjnr[jidx];
 658             j_coord_offsetA  = DIM*jnrA;
 659
 660             /* load j atom coordinates */
 661             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 662                                               &jx0,&jy0,&jz0);
 663
 664             /* Calculate displacement vector */
 665             dx00             = _mm_sub_pd(ix0,jx0);
 666             dy00             = _mm_sub_pd(iy0,jy0);
 667             dz00             = _mm_sub_pd(iz0,jz0);
 668             dx10             = _mm_sub_pd(ix1,jx0);
 669             dy10             = _mm_sub_pd(iy1,jy0);
 670             dz10             = _mm_sub_pd(iz1,jz0);
 671             dx20             = _mm_sub_pd(ix2,jx0);
 672             dy20             = _mm_sub_pd(iy2,jy0);
 673             dz20             = _mm_sub_pd(iz2,jz0);
 674
 675             /* Calculate squared distance and things based on it */
 676             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 677             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 678             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 679
 680             rinv00           = avx128fma_invsqrt_d(rsq00);
 681             rinv10           = avx128fma_invsqrt_d(rsq10);
 682             rinv20           = avx128fma_invsqrt_d(rsq20);
 683
 684             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 685             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 686             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 687
 688             /* Load parameters for j particles */
 689             jq0              = _mm_load_sd(charge+jnrA+0);
 690
 691             fjx0             = _mm_setzero_pd();
 692             fjy0             = _mm_setzero_pd();
 693             fjz0             = _mm_setzero_pd();
 694
 695             /**************************
 696              * CALCULATE INTERACTIONS *
 697              **************************/
 698
 699             /* Compute parameters for interactions between i and j atoms */
 700             qq00             = _mm_mul_pd(iq0,jq0);
 701
 702             /* REACTION-FIELD ELECTROSTATICS */
 703             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 704
 705             fscal            = felec;
 706
 707             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 708
 709             /* Update vectorial force */
 710             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 711             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 712             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 713
 714             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 715             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 716             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 717
 718             /**************************
 719              * CALCULATE INTERACTIONS *
 720              **************************/
 721
 722             /* Compute parameters for interactions between i and j atoms */
 723             qq10             = _mm_mul_pd(iq1,jq0);
 724
 725             /* REACTION-FIELD ELECTROSTATICS */
 726             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 727
 728             fscal            = felec;
 729
 730             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 731
 732             /* Update vectorial force */
 733             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 734             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 735             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 736
 737             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 738             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 739             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 740
 741             /**************************
 742              * CALCULATE INTERACTIONS *
 743              **************************/
 744
 745             /* Compute parameters for interactions between i and j atoms */
 746             qq20             = _mm_mul_pd(iq2,jq0);
 747
 748             /* REACTION-FIELD ELECTROSTATICS */
 749             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 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 }