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, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomW4P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            None
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwNone_GeomW4P1_VF_avx_128_fma_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwioffset3;
  87     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  91     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  92     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  93     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     __m128d          dummy_mask,cutoff_mask;
  96     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  97     __m128d          one     = _mm_set1_pd(1.0);
  98     __m128d          two     = _mm_set1_pd(2.0);
  99     x                = xx[0];
 100     f                = ff[0];
 101
 102     nri              = nlist->nri;
 103     iinr             = nlist->iinr;
 104     jindex           = nlist->jindex;
 105     jjnr             = nlist->jjnr;
 106     shiftidx         = nlist->shift;
 107     gid              = nlist->gid;
 108     shiftvec         = fr->shift_vec[0];
 109     fshift           = fr->fshift[0];
 110     facel            = _mm_set1_pd(fr->epsfac);
 111     charge           = mdatoms->chargeA;
 112     krf              = _mm_set1_pd(fr->ic->k_rf);
 113     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 114     crf              = _mm_set1_pd(fr->ic->c_rf);
 115
 116     /* Setup water-specific parameters */
 117     inr              = nlist->iinr[0];
 118     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 119     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 120     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 121
 122     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 123     rcutoff_scalar   = fr->rcoulomb;
 124     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 125     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 126
 127     /* Avoid stupid compiler warnings */
 128     jnrA = jnrB = 0;
 129     j_coord_offsetA = 0;
 130     j_coord_offsetB = 0;
 131
 132     outeriter        = 0;
 133     inneriter        = 0;
 134
 135     /* Start outer loop over neighborlists */
 136     for(iidx=0; iidx<nri; iidx++)
 137     {
 138         /* Load shift vector for this list */
 139         i_shift_offset   = DIM*shiftidx[iidx];
 140
 141         /* Load limits for loop over neighbors */
 142         j_index_start    = jindex[iidx];
 143         j_index_end      = jindex[iidx+1];
 144
 145         /* Get outer coordinate index */
 146         inr              = iinr[iidx];
 147         i_coord_offset   = DIM*inr;
 148
 149         /* Load i particle coords and add shift vector */
 150         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 151                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 152
 153         fix1             = _mm_setzero_pd();
 154         fiy1             = _mm_setzero_pd();
 155         fiz1             = _mm_setzero_pd();
 156         fix2             = _mm_setzero_pd();
 157         fiy2             = _mm_setzero_pd();
 158         fiz2             = _mm_setzero_pd();
 159         fix3             = _mm_setzero_pd();
 160         fiy3             = _mm_setzero_pd();
 161         fiz3             = _mm_setzero_pd();
 162
 163         /* Reset potential sums */
 164         velecsum         = _mm_setzero_pd();
 165
 166         /* Start inner kernel loop */
 167         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 168         {
 169
 170             /* Get j neighbor index, and coordinate index */
 171             jnrA             = jjnr[jidx];
 172             jnrB             = jjnr[jidx+1];
 173             j_coord_offsetA  = DIM*jnrA;
 174             j_coord_offsetB  = DIM*jnrB;
 175
 176             /* load j atom coordinates */
 177             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 178                                               &jx0,&jy0,&jz0);
 179
 180             /* Calculate displacement vector */
 181             dx10             = _mm_sub_pd(ix1,jx0);
 182             dy10             = _mm_sub_pd(iy1,jy0);
 183             dz10             = _mm_sub_pd(iz1,jz0);
 184             dx20             = _mm_sub_pd(ix2,jx0);
 185             dy20             = _mm_sub_pd(iy2,jy0);
 186             dz20             = _mm_sub_pd(iz2,jz0);
 187             dx30             = _mm_sub_pd(ix3,jx0);
 188             dy30             = _mm_sub_pd(iy3,jy0);
 189             dz30             = _mm_sub_pd(iz3,jz0);
 190
 191             /* Calculate squared distance and things based on it */
 192             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 193             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 194             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 195
 196             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 197             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 198             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 199
 200             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 201             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 202             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 203
 204             /* Load parameters for j particles */
 205             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 206
 207             fjx0             = _mm_setzero_pd();
 208             fjy0             = _mm_setzero_pd();
 209             fjz0             = _mm_setzero_pd();
 210
 211             /**************************
 212              * CALCULATE INTERACTIONS *
 213              **************************/
 214
 215             if (gmx_mm_any_lt(rsq10,rcutoff2))
 216             {
 217
 218             /* Compute parameters for interactions between i and j atoms */
 219             qq10             = _mm_mul_pd(iq1,jq0);
 220
 221             /* REACTION-FIELD ELECTROSTATICS */
 222             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 223             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 224
 225             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 226
 227             /* Update potential sum for this i atom from the interaction with this j atom. */
 228             velec            = _mm_and_pd(velec,cutoff_mask);
 229             velecsum         = _mm_add_pd(velecsum,velec);
 230
 231             fscal            = felec;
 232
 233             fscal            = _mm_and_pd(fscal,cutoff_mask);
 234
 235             /* Update vectorial force */
 236             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 237             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 238             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 239
 240             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 241             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 242             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 243
 244             }
 245
 246             /**************************
 247              * CALCULATE INTERACTIONS *
 248              **************************/
 249
 250             if (gmx_mm_any_lt(rsq20,rcutoff2))
 251             {
 252
 253             /* Compute parameters for interactions between i and j atoms */
 254             qq20             = _mm_mul_pd(iq2,jq0);
 255
 256             /* REACTION-FIELD ELECTROSTATICS */
 257             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 258             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 259
 260             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 261
 262             /* Update potential sum for this i atom from the interaction with this j atom. */
 263             velec            = _mm_and_pd(velec,cutoff_mask);
 264             velecsum         = _mm_add_pd(velecsum,velec);
 265
 266             fscal            = felec;
 267
 268             fscal            = _mm_and_pd(fscal,cutoff_mask);
 269
 270             /* Update vectorial force */
 271             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 272             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 273             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 274
 275             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 276             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 277             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 278
 279             }
 280
 281             /**************************
 282              * CALCULATE INTERACTIONS *
 283              **************************/
 284
 285             if (gmx_mm_any_lt(rsq30,rcutoff2))
 286             {
 287
 288             /* Compute parameters for interactions between i and j atoms */
 289             qq30             = _mm_mul_pd(iq3,jq0);
 290
 291             /* REACTION-FIELD ELECTROSTATICS */
 292             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_macc_pd(krf,rsq30,rinv30),crf));
 293             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 294
 295             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 296
 297             /* Update potential sum for this i atom from the interaction with this j atom. */
 298             velec            = _mm_and_pd(velec,cutoff_mask);
 299             velecsum         = _mm_add_pd(velecsum,velec);
 300
 301             fscal            = felec;
 302
 303             fscal            = _mm_and_pd(fscal,cutoff_mask);
 304
 305             /* Update vectorial force */
 306             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 307             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 308             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 309
 310             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 311             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 312             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 313
 314             }
 315
 316             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 317
 318             /* Inner loop uses 120 flops */
 319         }
 320
 321         if(jidx<j_index_end)
 322         {
 323
 324             jnrA             = jjnr[jidx];
 325             j_coord_offsetA  = DIM*jnrA;
 326
 327             /* load j atom coordinates */
 328             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 329                                               &jx0,&jy0,&jz0);
 330
 331             /* Calculate displacement vector */
 332             dx10             = _mm_sub_pd(ix1,jx0);
 333             dy10             = _mm_sub_pd(iy1,jy0);
 334             dz10             = _mm_sub_pd(iz1,jz0);
 335             dx20             = _mm_sub_pd(ix2,jx0);
 336             dy20             = _mm_sub_pd(iy2,jy0);
 337             dz20             = _mm_sub_pd(iz2,jz0);
 338             dx30             = _mm_sub_pd(ix3,jx0);
 339             dy30             = _mm_sub_pd(iy3,jy0);
 340             dz30             = _mm_sub_pd(iz3,jz0);
 341
 342             /* Calculate squared distance and things based on it */
 343             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 344             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 345             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 346
 347             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 348             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 349             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 350
 351             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 352             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 353             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 354
 355             /* Load parameters for j particles */
 356             jq0              = _mm_load_sd(charge+jnrA+0);
 357
 358             fjx0             = _mm_setzero_pd();
 359             fjy0             = _mm_setzero_pd();
 360             fjz0             = _mm_setzero_pd();
 361
 362             /**************************
 363              * CALCULATE INTERACTIONS *
 364              **************************/
 365
 366             if (gmx_mm_any_lt(rsq10,rcutoff2))
 367             {
 368
 369             /* Compute parameters for interactions between i and j atoms */
 370             qq10             = _mm_mul_pd(iq1,jq0);
 371
 372             /* REACTION-FIELD ELECTROSTATICS */
 373             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 374             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 375
 376             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 377
 378             /* Update potential sum for this i atom from the interaction with this j atom. */
 379             velec            = _mm_and_pd(velec,cutoff_mask);
 380             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 381             velecsum         = _mm_add_pd(velecsum,velec);
 382
 383             fscal            = felec;
 384
 385             fscal            = _mm_and_pd(fscal,cutoff_mask);
 386
 387             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 388
 389             /* Update vectorial force */
 390             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 391             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 392             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 393
 394             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 395             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 396             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 397
 398             }
 399
 400             /**************************
 401              * CALCULATE INTERACTIONS *
 402              **************************/
 403
 404             if (gmx_mm_any_lt(rsq20,rcutoff2))
 405             {
 406
 407             /* Compute parameters for interactions between i and j atoms */
 408             qq20             = _mm_mul_pd(iq2,jq0);
 409
 410             /* REACTION-FIELD ELECTROSTATICS */
 411             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 412             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 413
 414             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 415
 416             /* Update potential sum for this i atom from the interaction with this j atom. */
 417             velec            = _mm_and_pd(velec,cutoff_mask);
 418             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 419             velecsum         = _mm_add_pd(velecsum,velec);
 420
 421             fscal            = felec;
 422
 423             fscal            = _mm_and_pd(fscal,cutoff_mask);
 424
 425             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 426
 427             /* Update vectorial force */
 428             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 429             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 430             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 431
 432             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 433             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 434             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 435
 436             }
 437
 438             /**************************
 439              * CALCULATE INTERACTIONS *
 440              **************************/
 441
 442             if (gmx_mm_any_lt(rsq30,rcutoff2))
 443             {
 444
 445             /* Compute parameters for interactions between i and j atoms */
 446             qq30             = _mm_mul_pd(iq3,jq0);
 447
 448             /* REACTION-FIELD ELECTROSTATICS */
 449             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_macc_pd(krf,rsq30,rinv30),crf));
 450             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 451
 452             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 453
 454             /* Update potential sum for this i atom from the interaction with this j atom. */
 455             velec            = _mm_and_pd(velec,cutoff_mask);
 456             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 457             velecsum         = _mm_add_pd(velecsum,velec);
 458
 459             fscal            = felec;
 460
 461             fscal            = _mm_and_pd(fscal,cutoff_mask);
 462
 463             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 464
 465             /* Update vectorial force */
 466             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 467             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 468             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 469
 470             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 471             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 472             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 473
 474             }
 475
 476             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 477
 478             /* Inner loop uses 120 flops */
 479         }
 480
 481         /* End of innermost loop */
 482
 483         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 484                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 485
 486         ggid                        = gid[iidx];
 487         /* Update potential energies */
 488         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 489
 490         /* Increment number of inner iterations */
 491         inneriter                  += j_index_end - j_index_start;
 492
 493         /* Outer loop uses 19 flops */
 494     }
 495
 496     /* Increment number of outer iterations */
 497     outeriter        += nri;
 498
 499     /* Update outer/inner flops */
 500
 501     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*120);
 502 }
 503 /*
 504  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomW4P1_F_avx_128_fma_double
 505  * Electrostatics interaction: ReactionField
 506  * VdW interaction:            None
 507  * Geometry:                   Water4-Particle
 508  * Calculate force/pot:        Force
 509  */
 510 void
 511 nb_kernel_ElecRFCut_VdwNone_GeomW4P1_F_avx_128_fma_double
 512                     (t_nblist                    * gmx_restrict       nlist,
 513                      rvec                        * gmx_restrict          xx,
 514                      rvec                        * gmx_restrict          ff,
 515                      t_forcerec                  * gmx_restrict          fr,
 516                      t_mdatoms                   * gmx_restrict     mdatoms,
 517                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 518                      t_nrnb                      * gmx_restrict        nrnb)
 519 {
 520     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 521      * just 0 for non-waters.
 522      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 523      * jnr indices corresponding to data put in the four positions in the SIMD register.
 524      */
 525     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 526     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 527     int              jnrA,jnrB;
 528     int              j_coord_offsetA,j_coord_offsetB;
 529     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 530     real             rcutoff_scalar;
 531     real             *shiftvec,*fshift,*x,*f;
 532     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 533     int              vdwioffset1;
 534     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 535     int              vdwioffset2;
 536     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 537     int              vdwioffset3;
 538     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 539     int              vdwjidx0A,vdwjidx0B;
 540     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 541     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 542     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 543     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 544     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 545     real             *charge;
 546     __m128d          dummy_mask,cutoff_mask;
 547     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 548     __m128d          one     = _mm_set1_pd(1.0);
 549     __m128d          two     = _mm_set1_pd(2.0);
 550     x                = xx[0];
 551     f                = ff[0];
 552
 553     nri              = nlist->nri;
 554     iinr             = nlist->iinr;
 555     jindex           = nlist->jindex;
 556     jjnr             = nlist->jjnr;
 557     shiftidx         = nlist->shift;
 558     gid              = nlist->gid;
 559     shiftvec         = fr->shift_vec[0];
 560     fshift           = fr->fshift[0];
 561     facel            = _mm_set1_pd(fr->epsfac);
 562     charge           = mdatoms->chargeA;
 563     krf              = _mm_set1_pd(fr->ic->k_rf);
 564     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 565     crf              = _mm_set1_pd(fr->ic->c_rf);
 566
 567     /* Setup water-specific parameters */
 568     inr              = nlist->iinr[0];
 569     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 570     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 571     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 572
 573     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 574     rcutoff_scalar   = fr->rcoulomb;
 575     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 576     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 577
 578     /* Avoid stupid compiler warnings */
 579     jnrA = jnrB = 0;
 580     j_coord_offsetA = 0;
 581     j_coord_offsetB = 0;
 582
 583     outeriter        = 0;
 584     inneriter        = 0;
 585
 586     /* Start outer loop over neighborlists */
 587     for(iidx=0; iidx<nri; iidx++)
 588     {
 589         /* Load shift vector for this list */
 590         i_shift_offset   = DIM*shiftidx[iidx];
 591
 592         /* Load limits for loop over neighbors */
 593         j_index_start    = jindex[iidx];
 594         j_index_end      = jindex[iidx+1];
 595
 596         /* Get outer coordinate index */
 597         inr              = iinr[iidx];
 598         i_coord_offset   = DIM*inr;
 599
 600         /* Load i particle coords and add shift vector */
 601         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 602                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 603
 604         fix1             = _mm_setzero_pd();
 605         fiy1             = _mm_setzero_pd();
 606         fiz1             = _mm_setzero_pd();
 607         fix2             = _mm_setzero_pd();
 608         fiy2             = _mm_setzero_pd();
 609         fiz2             = _mm_setzero_pd();
 610         fix3             = _mm_setzero_pd();
 611         fiy3             = _mm_setzero_pd();
 612         fiz3             = _mm_setzero_pd();
 613
 614         /* Start inner kernel loop */
 615         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 616         {
 617
 618             /* Get j neighbor index, and coordinate index */
 619             jnrA             = jjnr[jidx];
 620             jnrB             = jjnr[jidx+1];
 621             j_coord_offsetA  = DIM*jnrA;
 622             j_coord_offsetB  = DIM*jnrB;
 623
 624             /* load j atom coordinates */
 625             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 626                                               &jx0,&jy0,&jz0);
 627
 628             /* Calculate displacement vector */
 629             dx10             = _mm_sub_pd(ix1,jx0);
 630             dy10             = _mm_sub_pd(iy1,jy0);
 631             dz10             = _mm_sub_pd(iz1,jz0);
 632             dx20             = _mm_sub_pd(ix2,jx0);
 633             dy20             = _mm_sub_pd(iy2,jy0);
 634             dz20             = _mm_sub_pd(iz2,jz0);
 635             dx30             = _mm_sub_pd(ix3,jx0);
 636             dy30             = _mm_sub_pd(iy3,jy0);
 637             dz30             = _mm_sub_pd(iz3,jz0);
 638
 639             /* Calculate squared distance and things based on it */
 640             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 641             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 642             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 643
 644             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 645             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 646             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 647
 648             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 649             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 650             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 651
 652             /* Load parameters for j particles */
 653             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 654
 655             fjx0             = _mm_setzero_pd();
 656             fjy0             = _mm_setzero_pd();
 657             fjz0             = _mm_setzero_pd();
 658
 659             /**************************
 660              * CALCULATE INTERACTIONS *
 661              **************************/
 662
 663             if (gmx_mm_any_lt(rsq10,rcutoff2))
 664             {
 665
 666             /* Compute parameters for interactions between i and j atoms */
 667             qq10             = _mm_mul_pd(iq1,jq0);
 668
 669             /* REACTION-FIELD ELECTROSTATICS */
 670             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 671
 672             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 673
 674             fscal            = felec;
 675
 676             fscal            = _mm_and_pd(fscal,cutoff_mask);
 677
 678             /* Update vectorial force */
 679             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 680             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 681             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 682
 683             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 684             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 685             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 686
 687             }
 688
 689             /**************************
 690              * CALCULATE INTERACTIONS *
 691              **************************/
 692
 693             if (gmx_mm_any_lt(rsq20,rcutoff2))
 694             {
 695
 696             /* Compute parameters for interactions between i and j atoms */
 697             qq20             = _mm_mul_pd(iq2,jq0);
 698
 699             /* REACTION-FIELD ELECTROSTATICS */
 700             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 701
 702             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 703
 704             fscal            = felec;
 705
 706             fscal            = _mm_and_pd(fscal,cutoff_mask);
 707
 708             /* Update vectorial force */
 709             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 710             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 711             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 712
 713             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 714             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 715             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 716
 717             }
 718
 719             /**************************
 720              * CALCULATE INTERACTIONS *
 721              **************************/
 722
 723             if (gmx_mm_any_lt(rsq30,rcutoff2))
 724             {
 725
 726             /* Compute parameters for interactions between i and j atoms */
 727             qq30             = _mm_mul_pd(iq3,jq0);
 728
 729             /* REACTION-FIELD ELECTROSTATICS */
 730             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 731
 732             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 733
 734             fscal            = felec;
 735
 736             fscal            = _mm_and_pd(fscal,cutoff_mask);
 737
 738             /* Update vectorial force */
 739             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 740             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 741             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 742
 743             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 744             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 745             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 746
 747             }
 748
 749             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 750
 751             /* Inner loop uses 102 flops */
 752         }
 753
 754         if(jidx<j_index_end)
 755         {
 756
 757             jnrA             = jjnr[jidx];
 758             j_coord_offsetA  = DIM*jnrA;
 759
 760             /* load j atom coordinates */
 761             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 762                                               &jx0,&jy0,&jz0);
 763
 764             /* Calculate displacement vector */
 765             dx10             = _mm_sub_pd(ix1,jx0);
 766             dy10             = _mm_sub_pd(iy1,jy0);
 767             dz10             = _mm_sub_pd(iz1,jz0);
 768             dx20             = _mm_sub_pd(ix2,jx0);
 769             dy20             = _mm_sub_pd(iy2,jy0);
 770             dz20             = _mm_sub_pd(iz2,jz0);
 771             dx30             = _mm_sub_pd(ix3,jx0);
 772             dy30             = _mm_sub_pd(iy3,jy0);
 773             dz30             = _mm_sub_pd(iz3,jz0);
 774
 775             /* Calculate squared distance and things based on it */
 776             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 777             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 778             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 779
 780             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 781             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 782             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 783
 784             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 785             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 786             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 787
 788             /* Load parameters for j particles */
 789             jq0              = _mm_load_sd(charge+jnrA+0);
 790
 791             fjx0             = _mm_setzero_pd();
 792             fjy0             = _mm_setzero_pd();
 793             fjz0             = _mm_setzero_pd();
 794
 795             /**************************
 796              * CALCULATE INTERACTIONS *
 797              **************************/
 798
 799             if (gmx_mm_any_lt(rsq10,rcutoff2))
 800             {
 801
 802             /* Compute parameters for interactions between i and j atoms */
 803             qq10             = _mm_mul_pd(iq1,jq0);
 804
 805             /* REACTION-FIELD ELECTROSTATICS */
 806             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 807
 808             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 809
 810             fscal            = felec;
 811
 812             fscal            = _mm_and_pd(fscal,cutoff_mask);
 813
 814             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 815
 816             /* Update vectorial force */
 817             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 818             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 819             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 820
 821             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 822             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 823             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 824
 825             }
 826
 827             /**************************
 828              * CALCULATE INTERACTIONS *
 829              **************************/
 830
 831             if (gmx_mm_any_lt(rsq20,rcutoff2))
 832             {
 833
 834             /* Compute parameters for interactions between i and j atoms */
 835             qq20             = _mm_mul_pd(iq2,jq0);
 836
 837             /* REACTION-FIELD ELECTROSTATICS */
 838             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 839
 840             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 841
 842             fscal            = felec;
 843
 844             fscal            = _mm_and_pd(fscal,cutoff_mask);
 845
 846             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 847
 848             /* Update vectorial force */
 849             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 850             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 851             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 852
 853             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 854             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 855             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 856
 857             }
 858
 859             /**************************
 860              * CALCULATE INTERACTIONS *
 861              **************************/
 862
 863             if (gmx_mm_any_lt(rsq30,rcutoff2))
 864             {
 865
 866             /* Compute parameters for interactions between i and j atoms */
 867             qq30             = _mm_mul_pd(iq3,jq0);
 868
 869             /* REACTION-FIELD ELECTROSTATICS */
 870             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 871
 872             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 873
 874             fscal            = felec;
 875
 876             fscal            = _mm_and_pd(fscal,cutoff_mask);
 877
 878             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 879
 880             /* Update vectorial force */
 881             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 882             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 883             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 884
 885             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 886             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 887             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 888
 889             }
 890
 891             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 892
 893             /* Inner loop uses 102 flops */
 894         }
 895
 896         /* End of innermost loop */
 897
 898         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 899                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 900
 901         /* Increment number of inner iterations */
 902         inneriter                  += j_index_end - j_index_start;
 903
 904         /* Outer loop uses 18 flops */
 905     }
 906
 907     /* Increment number of outer iterations */
 908     outeriter        += nri;
 909
 910     /* Update outer/inner flops */
 911
 912     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*102);
 913 }