src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecRFCut_VdwNone_GeomW4P1_avx_128_fma_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  48 #include "kernelutil_x86_avx_128_fma_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomW4P1_VF_avx_128_fma_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            None
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwNone_GeomW4P1_VF_avx_128_fma_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset1;
  81     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  82     int              vdwioffset2;
  83     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  84     int              vdwioffset3;
  85     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  86     int              vdwjidx0A,vdwjidx0B;
  87     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  89     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  90     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  91     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     __m128d          dummy_mask,cutoff_mask;
  94     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  95     __m128d          one     = _mm_set1_pd(1.0);
  96     __m128d          two     = _mm_set1_pd(2.0);
  97     x                = xx[0];
  98     f                = ff[0];
  99
 100     nri              = nlist->nri;
 101     iinr             = nlist->iinr;
 102     jindex           = nlist->jindex;
 103     jjnr             = nlist->jjnr;
 104     shiftidx         = nlist->shift;
 105     gid              = nlist->gid;
 106     shiftvec         = fr->shift_vec[0];
 107     fshift           = fr->fshift[0];
 108     facel            = _mm_set1_pd(fr->epsfac);
 109     charge           = mdatoms->chargeA;
 110     krf              = _mm_set1_pd(fr->ic->k_rf);
 111     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 112     crf              = _mm_set1_pd(fr->ic->c_rf);
 113
 114     /* Setup water-specific parameters */
 115     inr              = nlist->iinr[0];
 116     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 117     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 118     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 119
 120     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 121     rcutoff_scalar   = fr->rcoulomb;
 122     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 123     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 124
 125     /* Avoid stupid compiler warnings */
 126     jnrA = jnrB = 0;
 127     j_coord_offsetA = 0;
 128     j_coord_offsetB = 0;
 129
 130     outeriter        = 0;
 131     inneriter        = 0;
 132
 133     /* Start outer loop over neighborlists */
 134     for(iidx=0; iidx<nri; iidx++)
 135     {
 136         /* Load shift vector for this list */
 137         i_shift_offset   = DIM*shiftidx[iidx];
 138
 139         /* Load limits for loop over neighbors */
 140         j_index_start    = jindex[iidx];
 141         j_index_end      = jindex[iidx+1];
 142
 143         /* Get outer coordinate index */
 144         inr              = iinr[iidx];
 145         i_coord_offset   = DIM*inr;
 146
 147         /* Load i particle coords and add shift vector */
 148         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 149                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 150
 151         fix1             = _mm_setzero_pd();
 152         fiy1             = _mm_setzero_pd();
 153         fiz1             = _mm_setzero_pd();
 154         fix2             = _mm_setzero_pd();
 155         fiy2             = _mm_setzero_pd();
 156         fiz2             = _mm_setzero_pd();
 157         fix3             = _mm_setzero_pd();
 158         fiy3             = _mm_setzero_pd();
 159         fiz3             = _mm_setzero_pd();
 160
 161         /* Reset potential sums */
 162         velecsum         = _mm_setzero_pd();
 163
 164         /* Start inner kernel loop */
 165         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 166         {
 167
 168             /* Get j neighbor index, and coordinate index */
 169             jnrA             = jjnr[jidx];
 170             jnrB             = jjnr[jidx+1];
 171             j_coord_offsetA  = DIM*jnrA;
 172             j_coord_offsetB  = DIM*jnrB;
 173
 174             /* load j atom coordinates */
 175             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 176                                               &jx0,&jy0,&jz0);
 177
 178             /* Calculate displacement vector */
 179             dx10             = _mm_sub_pd(ix1,jx0);
 180             dy10             = _mm_sub_pd(iy1,jy0);
 181             dz10             = _mm_sub_pd(iz1,jz0);
 182             dx20             = _mm_sub_pd(ix2,jx0);
 183             dy20             = _mm_sub_pd(iy2,jy0);
 184             dz20             = _mm_sub_pd(iz2,jz0);
 185             dx30             = _mm_sub_pd(ix3,jx0);
 186             dy30             = _mm_sub_pd(iy3,jy0);
 187             dz30             = _mm_sub_pd(iz3,jz0);
 188
 189             /* Calculate squared distance and things based on it */
 190             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 191             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 192             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 193
 194             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 195             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 196             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 197
 198             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 199             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 200             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 201
 202             /* Load parameters for j particles */
 203             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 204
 205             fjx0             = _mm_setzero_pd();
 206             fjy0             = _mm_setzero_pd();
 207             fjz0             = _mm_setzero_pd();
 208
 209             /**************************
 210              * CALCULATE INTERACTIONS *
 211              **************************/
 212
 213             if (gmx_mm_any_lt(rsq10,rcutoff2))
 214             {
 215
 216             /* Compute parameters for interactions between i and j atoms */
 217             qq10             = _mm_mul_pd(iq1,jq0);
 218
 219             /* REACTION-FIELD ELECTROSTATICS */
 220             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 221             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 222
 223             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 224
 225             /* Update potential sum for this i atom from the interaction with this j atom. */
 226             velec            = _mm_and_pd(velec,cutoff_mask);
 227             velecsum         = _mm_add_pd(velecsum,velec);
 228
 229             fscal            = felec;
 230
 231             fscal            = _mm_and_pd(fscal,cutoff_mask);
 232
 233             /* Update vectorial force */
 234             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 235             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 236             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 237
 238             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 239             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 240             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 241
 242             }
 243
 244             /**************************
 245              * CALCULATE INTERACTIONS *
 246              **************************/
 247
 248             if (gmx_mm_any_lt(rsq20,rcutoff2))
 249             {
 250
 251             /* Compute parameters for interactions between i and j atoms */
 252             qq20             = _mm_mul_pd(iq2,jq0);
 253
 254             /* REACTION-FIELD ELECTROSTATICS */
 255             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 256             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 257
 258             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 259
 260             /* Update potential sum for this i atom from the interaction with this j atom. */
 261             velec            = _mm_and_pd(velec,cutoff_mask);
 262             velecsum         = _mm_add_pd(velecsum,velec);
 263
 264             fscal            = felec;
 265
 266             fscal            = _mm_and_pd(fscal,cutoff_mask);
 267
 268             /* Update vectorial force */
 269             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 270             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 271             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 272
 273             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 274             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 275             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 276
 277             }
 278
 279             /**************************
 280              * CALCULATE INTERACTIONS *
 281              **************************/
 282
 283             if (gmx_mm_any_lt(rsq30,rcutoff2))
 284             {
 285
 286             /* Compute parameters for interactions between i and j atoms */
 287             qq30             = _mm_mul_pd(iq3,jq0);
 288
 289             /* REACTION-FIELD ELECTROSTATICS */
 290             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_macc_pd(krf,rsq30,rinv30),crf));
 291             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 292
 293             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 294
 295             /* Update potential sum for this i atom from the interaction with this j atom. */
 296             velec            = _mm_and_pd(velec,cutoff_mask);
 297             velecsum         = _mm_add_pd(velecsum,velec);
 298
 299             fscal            = felec;
 300
 301             fscal            = _mm_and_pd(fscal,cutoff_mask);
 302
 303             /* Update vectorial force */
 304             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 305             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 306             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 307
 308             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 309             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 310             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 311
 312             }
 313
 314             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 315
 316             /* Inner loop uses 120 flops */
 317         }
 318
 319         if(jidx<j_index_end)
 320         {
 321
 322             jnrA             = jjnr[jidx];
 323             j_coord_offsetA  = DIM*jnrA;
 324
 325             /* load j atom coordinates */
 326             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 327                                               &jx0,&jy0,&jz0);
 328
 329             /* Calculate displacement vector */
 330             dx10             = _mm_sub_pd(ix1,jx0);
 331             dy10             = _mm_sub_pd(iy1,jy0);
 332             dz10             = _mm_sub_pd(iz1,jz0);
 333             dx20             = _mm_sub_pd(ix2,jx0);
 334             dy20             = _mm_sub_pd(iy2,jy0);
 335             dz20             = _mm_sub_pd(iz2,jz0);
 336             dx30             = _mm_sub_pd(ix3,jx0);
 337             dy30             = _mm_sub_pd(iy3,jy0);
 338             dz30             = _mm_sub_pd(iz3,jz0);
 339
 340             /* Calculate squared distance and things based on it */
 341             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 342             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 343             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 344
 345             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 346             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 347             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 348
 349             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 350             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 351             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 352
 353             /* Load parameters for j particles */
 354             jq0              = _mm_load_sd(charge+jnrA+0);
 355
 356             fjx0             = _mm_setzero_pd();
 357             fjy0             = _mm_setzero_pd();
 358             fjz0             = _mm_setzero_pd();
 359
 360             /**************************
 361              * CALCULATE INTERACTIONS *
 362              **************************/
 363
 364             if (gmx_mm_any_lt(rsq10,rcutoff2))
 365             {
 366
 367             /* Compute parameters for interactions between i and j atoms */
 368             qq10             = _mm_mul_pd(iq1,jq0);
 369
 370             /* REACTION-FIELD ELECTROSTATICS */
 371             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 372             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 373
 374             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 375
 376             /* Update potential sum for this i atom from the interaction with this j atom. */
 377             velec            = _mm_and_pd(velec,cutoff_mask);
 378             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 379             velecsum         = _mm_add_pd(velecsum,velec);
 380
 381             fscal            = felec;
 382
 383             fscal            = _mm_and_pd(fscal,cutoff_mask);
 384
 385             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 386
 387             /* Update vectorial force */
 388             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 389             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 390             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 391
 392             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 393             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 394             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 395
 396             }
 397
 398             /**************************
 399              * CALCULATE INTERACTIONS *
 400              **************************/
 401
 402             if (gmx_mm_any_lt(rsq20,rcutoff2))
 403             {
 404
 405             /* Compute parameters for interactions between i and j atoms */
 406             qq20             = _mm_mul_pd(iq2,jq0);
 407
 408             /* REACTION-FIELD ELECTROSTATICS */
 409             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 410             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 411
 412             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 413
 414             /* Update potential sum for this i atom from the interaction with this j atom. */
 415             velec            = _mm_and_pd(velec,cutoff_mask);
 416             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 417             velecsum         = _mm_add_pd(velecsum,velec);
 418
 419             fscal            = felec;
 420
 421             fscal            = _mm_and_pd(fscal,cutoff_mask);
 422
 423             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 424
 425             /* Update vectorial force */
 426             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 427             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 428             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 429
 430             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 431             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 432             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 433
 434             }
 435
 436             /**************************
 437              * CALCULATE INTERACTIONS *
 438              **************************/
 439
 440             if (gmx_mm_any_lt(rsq30,rcutoff2))
 441             {
 442
 443             /* Compute parameters for interactions between i and j atoms */
 444             qq30             = _mm_mul_pd(iq3,jq0);
 445
 446             /* REACTION-FIELD ELECTROSTATICS */
 447             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_macc_pd(krf,rsq30,rinv30),crf));
 448             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 449
 450             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 451
 452             /* Update potential sum for this i atom from the interaction with this j atom. */
 453             velec            = _mm_and_pd(velec,cutoff_mask);
 454             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 455             velecsum         = _mm_add_pd(velecsum,velec);
 456
 457             fscal            = felec;
 458
 459             fscal            = _mm_and_pd(fscal,cutoff_mask);
 460
 461             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 462
 463             /* Update vectorial force */
 464             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 465             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 466             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 467
 468             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 469             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 470             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 471
 472             }
 473
 474             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 475
 476             /* Inner loop uses 120 flops */
 477         }
 478
 479         /* End of innermost loop */
 480
 481         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 482                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 483
 484         ggid                        = gid[iidx];
 485         /* Update potential energies */
 486         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 487
 488         /* Increment number of inner iterations */
 489         inneriter                  += j_index_end - j_index_start;
 490
 491         /* Outer loop uses 19 flops */
 492     }
 493
 494     /* Increment number of outer iterations */
 495     outeriter        += nri;
 496
 497     /* Update outer/inner flops */
 498
 499     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*120);
 500 }
 501 /*
 502  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomW4P1_F_avx_128_fma_double
 503  * Electrostatics interaction: ReactionField
 504  * VdW interaction:            None
 505  * Geometry:                   Water4-Particle
 506  * Calculate force/pot:        Force
 507  */
 508 void
 509 nb_kernel_ElecRFCut_VdwNone_GeomW4P1_F_avx_128_fma_double
 510                     (t_nblist                    * gmx_restrict       nlist,
 511                      rvec                        * gmx_restrict          xx,
 512                      rvec                        * gmx_restrict          ff,
 513                      t_forcerec                  * gmx_restrict          fr,
 514                      t_mdatoms                   * gmx_restrict     mdatoms,
 515                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 516                      t_nrnb                      * gmx_restrict        nrnb)
 517 {
 518     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 519      * just 0 for non-waters.
 520      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 521      * jnr indices corresponding to data put in the four positions in the SIMD register.
 522      */
 523     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 524     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 525     int              jnrA,jnrB;
 526     int              j_coord_offsetA,j_coord_offsetB;
 527     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 528     real             rcutoff_scalar;
 529     real             *shiftvec,*fshift,*x,*f;
 530     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 531     int              vdwioffset1;
 532     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 533     int              vdwioffset2;
 534     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 535     int              vdwioffset3;
 536     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 537     int              vdwjidx0A,vdwjidx0B;
 538     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 539     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 540     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 541     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 542     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 543     real             *charge;
 544     __m128d          dummy_mask,cutoff_mask;
 545     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 546     __m128d          one     = _mm_set1_pd(1.0);
 547     __m128d          two     = _mm_set1_pd(2.0);
 548     x                = xx[0];
 549     f                = ff[0];
 550
 551     nri              = nlist->nri;
 552     iinr             = nlist->iinr;
 553     jindex           = nlist->jindex;
 554     jjnr             = nlist->jjnr;
 555     shiftidx         = nlist->shift;
 556     gid              = nlist->gid;
 557     shiftvec         = fr->shift_vec[0];
 558     fshift           = fr->fshift[0];
 559     facel            = _mm_set1_pd(fr->epsfac);
 560     charge           = mdatoms->chargeA;
 561     krf              = _mm_set1_pd(fr->ic->k_rf);
 562     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 563     crf              = _mm_set1_pd(fr->ic->c_rf);
 564
 565     /* Setup water-specific parameters */
 566     inr              = nlist->iinr[0];
 567     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 568     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 569     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 570
 571     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 572     rcutoff_scalar   = fr->rcoulomb;
 573     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 574     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 575
 576     /* Avoid stupid compiler warnings */
 577     jnrA = jnrB = 0;
 578     j_coord_offsetA = 0;
 579     j_coord_offsetB = 0;
 580
 581     outeriter        = 0;
 582     inneriter        = 0;
 583
 584     /* Start outer loop over neighborlists */
 585     for(iidx=0; iidx<nri; iidx++)
 586     {
 587         /* Load shift vector for this list */
 588         i_shift_offset   = DIM*shiftidx[iidx];
 589
 590         /* Load limits for loop over neighbors */
 591         j_index_start    = jindex[iidx];
 592         j_index_end      = jindex[iidx+1];
 593
 594         /* Get outer coordinate index */
 595         inr              = iinr[iidx];
 596         i_coord_offset   = DIM*inr;
 597
 598         /* Load i particle coords and add shift vector */
 599         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 600                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 601
 602         fix1             = _mm_setzero_pd();
 603         fiy1             = _mm_setzero_pd();
 604         fiz1             = _mm_setzero_pd();
 605         fix2             = _mm_setzero_pd();
 606         fiy2             = _mm_setzero_pd();
 607         fiz2             = _mm_setzero_pd();
 608         fix3             = _mm_setzero_pd();
 609         fiy3             = _mm_setzero_pd();
 610         fiz3             = _mm_setzero_pd();
 611
 612         /* Start inner kernel loop */
 613         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 614         {
 615
 616             /* Get j neighbor index, and coordinate index */
 617             jnrA             = jjnr[jidx];
 618             jnrB             = jjnr[jidx+1];
 619             j_coord_offsetA  = DIM*jnrA;
 620             j_coord_offsetB  = DIM*jnrB;
 621
 622             /* load j atom coordinates */
 623             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 624                                               &jx0,&jy0,&jz0);
 625
 626             /* Calculate displacement vector */
 627             dx10             = _mm_sub_pd(ix1,jx0);
 628             dy10             = _mm_sub_pd(iy1,jy0);
 629             dz10             = _mm_sub_pd(iz1,jz0);
 630             dx20             = _mm_sub_pd(ix2,jx0);
 631             dy20             = _mm_sub_pd(iy2,jy0);
 632             dz20             = _mm_sub_pd(iz2,jz0);
 633             dx30             = _mm_sub_pd(ix3,jx0);
 634             dy30             = _mm_sub_pd(iy3,jy0);
 635             dz30             = _mm_sub_pd(iz3,jz0);
 636
 637             /* Calculate squared distance and things based on it */
 638             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 639             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 640             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 641
 642             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 643             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 644             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 645
 646             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 647             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 648             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 649
 650             /* Load parameters for j particles */
 651             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 652
 653             fjx0             = _mm_setzero_pd();
 654             fjy0             = _mm_setzero_pd();
 655             fjz0             = _mm_setzero_pd();
 656
 657             /**************************
 658              * CALCULATE INTERACTIONS *
 659              **************************/
 660
 661             if (gmx_mm_any_lt(rsq10,rcutoff2))
 662             {
 663
 664             /* Compute parameters for interactions between i and j atoms */
 665             qq10             = _mm_mul_pd(iq1,jq0);
 666
 667             /* REACTION-FIELD ELECTROSTATICS */
 668             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 669
 670             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 671
 672             fscal            = felec;
 673
 674             fscal            = _mm_and_pd(fscal,cutoff_mask);
 675
 676             /* Update vectorial force */
 677             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 678             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 679             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 680
 681             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 682             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 683             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 684
 685             }
 686
 687             /**************************
 688              * CALCULATE INTERACTIONS *
 689              **************************/
 690
 691             if (gmx_mm_any_lt(rsq20,rcutoff2))
 692             {
 693
 694             /* Compute parameters for interactions between i and j atoms */
 695             qq20             = _mm_mul_pd(iq2,jq0);
 696
 697             /* REACTION-FIELD ELECTROSTATICS */
 698             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 699
 700             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 701
 702             fscal            = felec;
 703
 704             fscal            = _mm_and_pd(fscal,cutoff_mask);
 705
 706             /* Update vectorial force */
 707             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 708             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 709             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 710
 711             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 712             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 713             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 714
 715             }
 716
 717             /**************************
 718              * CALCULATE INTERACTIONS *
 719              **************************/
 720
 721             if (gmx_mm_any_lt(rsq30,rcutoff2))
 722             {
 723
 724             /* Compute parameters for interactions between i and j atoms */
 725             qq30             = _mm_mul_pd(iq3,jq0);
 726
 727             /* REACTION-FIELD ELECTROSTATICS */
 728             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 729
 730             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 731
 732             fscal            = felec;
 733
 734             fscal            = _mm_and_pd(fscal,cutoff_mask);
 735
 736             /* Update vectorial force */
 737             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 738             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 739             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 740
 741             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 742             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 743             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 744
 745             }
 746
 747             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 748
 749             /* Inner loop uses 102 flops */
 750         }
 751
 752         if(jidx<j_index_end)
 753         {
 754
 755             jnrA             = jjnr[jidx];
 756             j_coord_offsetA  = DIM*jnrA;
 757
 758             /* load j atom coordinates */
 759             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 760                                               &jx0,&jy0,&jz0);
 761
 762             /* Calculate displacement vector */
 763             dx10             = _mm_sub_pd(ix1,jx0);
 764             dy10             = _mm_sub_pd(iy1,jy0);
 765             dz10             = _mm_sub_pd(iz1,jz0);
 766             dx20             = _mm_sub_pd(ix2,jx0);
 767             dy20             = _mm_sub_pd(iy2,jy0);
 768             dz20             = _mm_sub_pd(iz2,jz0);
 769             dx30             = _mm_sub_pd(ix3,jx0);
 770             dy30             = _mm_sub_pd(iy3,jy0);
 771             dz30             = _mm_sub_pd(iz3,jz0);
 772
 773             /* Calculate squared distance and things based on it */
 774             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 775             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 776             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 777
 778             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 779             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 780             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 781
 782             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 783             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 784             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 785
 786             /* Load parameters for j particles */
 787             jq0              = _mm_load_sd(charge+jnrA+0);
 788
 789             fjx0             = _mm_setzero_pd();
 790             fjy0             = _mm_setzero_pd();
 791             fjz0             = _mm_setzero_pd();
 792
 793             /**************************
 794              * CALCULATE INTERACTIONS *
 795              **************************/
 796
 797             if (gmx_mm_any_lt(rsq10,rcutoff2))
 798             {
 799
 800             /* Compute parameters for interactions between i and j atoms */
 801             qq10             = _mm_mul_pd(iq1,jq0);
 802
 803             /* REACTION-FIELD ELECTROSTATICS */
 804             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 805
 806             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 807
 808             fscal            = felec;
 809
 810             fscal            = _mm_and_pd(fscal,cutoff_mask);
 811
 812             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 813
 814             /* Update vectorial force */
 815             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 816             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 817             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 818
 819             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 820             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 821             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 822
 823             }
 824
 825             /**************************
 826              * CALCULATE INTERACTIONS *
 827              **************************/
 828
 829             if (gmx_mm_any_lt(rsq20,rcutoff2))
 830             {
 831
 832             /* Compute parameters for interactions between i and j atoms */
 833             qq20             = _mm_mul_pd(iq2,jq0);
 834
 835             /* REACTION-FIELD ELECTROSTATICS */
 836             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 837
 838             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 839
 840             fscal            = felec;
 841
 842             fscal            = _mm_and_pd(fscal,cutoff_mask);
 843
 844             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 845
 846             /* Update vectorial force */
 847             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 848             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 849             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 850
 851             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 852             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 853             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 854
 855             }
 856
 857             /**************************
 858              * CALCULATE INTERACTIONS *
 859              **************************/
 860
 861             if (gmx_mm_any_lt(rsq30,rcutoff2))
 862             {
 863
 864             /* Compute parameters for interactions between i and j atoms */
 865             qq30             = _mm_mul_pd(iq3,jq0);
 866
 867             /* REACTION-FIELD ELECTROSTATICS */
 868             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 869
 870             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 871
 872             fscal            = felec;
 873
 874             fscal            = _mm_and_pd(fscal,cutoff_mask);
 875
 876             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 877
 878             /* Update vectorial force */
 879             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 880             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 881             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 882
 883             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 884             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 885             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 886
 887             }
 888
 889             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 890
 891             /* Inner loop uses 102 flops */
 892         }
 893
 894         /* End of innermost loop */
 895
 896         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 897                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 898
 899         /* Increment number of inner iterations */
 900         inneriter                  += j_index_end - j_index_start;
 901
 902         /* Outer loop uses 18 flops */
 903     }
 904
 905     /* Increment number of outer iterations */
 906     outeriter        += nri;
 907
 908     /* Update outer/inner flops */
 909
 910     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*102);
 911 }