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) 2012,2013,2014,2015,2017, 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_ElecRFCut_VdwNone_GeomW4P1_VF_avx_128_fma_double
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            None
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRFCut_VdwNone_GeomW4P1_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              vdwioffset1;
  80     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  81     int              vdwioffset2;
  82     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  83     int              vdwioffset3;
  84     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  85     int              vdwjidx0A,vdwjidx0B;
  86     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  88     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  89     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  90     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     __m128d          dummy_mask,cutoff_mask;
  93     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  94     __m128d          one     = _mm_set1_pd(1.0);
  95     __m128d          two     = _mm_set1_pd(2.0);
  96     x                = xx[0];
  97     f                = ff[0];
  98
  99     nri              = nlist->nri;
 100     iinr             = nlist->iinr;
 101     jindex           = nlist->jindex;
 102     jjnr             = nlist->jjnr;
 103     shiftidx         = nlist->shift;
 104     gid              = nlist->gid;
 105     shiftvec         = fr->shift_vec[0];
 106     fshift           = fr->fshift[0];
 107     facel            = _mm_set1_pd(fr->ic->epsfac);
 108     charge           = mdatoms->chargeA;
 109     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     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 116     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 117     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 118
 119     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 120     rcutoff_scalar   = fr->ic->rcoulomb;
 121     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 122     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 123
 124     /* Avoid stupid compiler warnings */
 125     jnrA = jnrB = 0;
 126     j_coord_offsetA = 0;
 127     j_coord_offsetB = 0;
 128
 129     outeriter        = 0;
 130     inneriter        = 0;
 131
 132     /* Start outer loop over neighborlists */
 133     for(iidx=0; iidx<nri; iidx++)
 134     {
 135         /* Load shift vector for this list */
 136         i_shift_offset   = DIM*shiftidx[iidx];
 137
 138         /* Load limits for loop over neighbors */
 139         j_index_start    = jindex[iidx];
 140         j_index_end      = jindex[iidx+1];
 141
 142         /* Get outer coordinate index */
 143         inr              = iinr[iidx];
 144         i_coord_offset   = DIM*inr;
 145
 146         /* Load i particle coords and add shift vector */
 147         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 148                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 149
 150         fix1             = _mm_setzero_pd();
 151         fiy1             = _mm_setzero_pd();
 152         fiz1             = _mm_setzero_pd();
 153         fix2             = _mm_setzero_pd();
 154         fiy2             = _mm_setzero_pd();
 155         fiz2             = _mm_setzero_pd();
 156         fix3             = _mm_setzero_pd();
 157         fiy3             = _mm_setzero_pd();
 158         fiz3             = _mm_setzero_pd();
 159
 160         /* Reset potential sums */
 161         velecsum         = _mm_setzero_pd();
 162
 163         /* Start inner kernel loop */
 164         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 165         {
 166
 167             /* Get j neighbor index, and coordinate index */
 168             jnrA             = jjnr[jidx];
 169             jnrB             = jjnr[jidx+1];
 170             j_coord_offsetA  = DIM*jnrA;
 171             j_coord_offsetB  = DIM*jnrB;
 172
 173             /* load j atom coordinates */
 174             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 175                                               &jx0,&jy0,&jz0);
 176
 177             /* Calculate displacement vector */
 178             dx10             = _mm_sub_pd(ix1,jx0);
 179             dy10             = _mm_sub_pd(iy1,jy0);
 180             dz10             = _mm_sub_pd(iz1,jz0);
 181             dx20             = _mm_sub_pd(ix2,jx0);
 182             dy20             = _mm_sub_pd(iy2,jy0);
 183             dz20             = _mm_sub_pd(iz2,jz0);
 184             dx30             = _mm_sub_pd(ix3,jx0);
 185             dy30             = _mm_sub_pd(iy3,jy0);
 186             dz30             = _mm_sub_pd(iz3,jz0);
 187
 188             /* Calculate squared distance and things based on it */
 189             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 190             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 191             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 192
 193             rinv10           = avx128fma_invsqrt_d(rsq10);
 194             rinv20           = avx128fma_invsqrt_d(rsq20);
 195             rinv30           = avx128fma_invsqrt_d(rsq30);
 196
 197             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 198             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 199             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 200
 201             /* Load parameters for j particles */
 202             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 203
 204             fjx0             = _mm_setzero_pd();
 205             fjy0             = _mm_setzero_pd();
 206             fjz0             = _mm_setzero_pd();
 207
 208             /**************************
 209              * CALCULATE INTERACTIONS *
 210              **************************/
 211
 212             if (gmx_mm_any_lt(rsq10,rcutoff2))
 213             {
 214
 215             /* Compute parameters for interactions between i and j atoms */
 216             qq10             = _mm_mul_pd(iq1,jq0);
 217
 218             /* REACTION-FIELD ELECTROSTATICS */
 219             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 220             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 221
 222             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 223
 224             /* Update potential sum for this i atom from the interaction with this j atom. */
 225             velec            = _mm_and_pd(velec,cutoff_mask);
 226             velecsum         = _mm_add_pd(velecsum,velec);
 227
 228             fscal            = felec;
 229
 230             fscal            = _mm_and_pd(fscal,cutoff_mask);
 231
 232             /* Update vectorial force */
 233             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 234             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 235             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 236
 237             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 238             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 239             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 240
 241             }
 242
 243             /**************************
 244              * CALCULATE INTERACTIONS *
 245              **************************/
 246
 247             if (gmx_mm_any_lt(rsq20,rcutoff2))
 248             {
 249
 250             /* Compute parameters for interactions between i and j atoms */
 251             qq20             = _mm_mul_pd(iq2,jq0);
 252
 253             /* REACTION-FIELD ELECTROSTATICS */
 254             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 255             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 256
 257             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 258
 259             /* Update potential sum for this i atom from the interaction with this j atom. */
 260             velec            = _mm_and_pd(velec,cutoff_mask);
 261             velecsum         = _mm_add_pd(velecsum,velec);
 262
 263             fscal            = felec;
 264
 265             fscal            = _mm_and_pd(fscal,cutoff_mask);
 266
 267             /* Update vectorial force */
 268             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 269             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 270             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 271
 272             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 273             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 274             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 275
 276             }
 277
 278             /**************************
 279              * CALCULATE INTERACTIONS *
 280              **************************/
 281
 282             if (gmx_mm_any_lt(rsq30,rcutoff2))
 283             {
 284
 285             /* Compute parameters for interactions between i and j atoms */
 286             qq30             = _mm_mul_pd(iq3,jq0);
 287
 288             /* REACTION-FIELD ELECTROSTATICS */
 289             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_macc_pd(krf,rsq30,rinv30),crf));
 290             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 291
 292             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 293
 294             /* Update potential sum for this i atom from the interaction with this j atom. */
 295             velec            = _mm_and_pd(velec,cutoff_mask);
 296             velecsum         = _mm_add_pd(velecsum,velec);
 297
 298             fscal            = felec;
 299
 300             fscal            = _mm_and_pd(fscal,cutoff_mask);
 301
 302             /* Update vectorial force */
 303             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 304             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 305             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 306
 307             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 308             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 309             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 310
 311             }
 312
 313             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 314
 315             /* Inner loop uses 120 flops */
 316         }
 317
 318         if(jidx<j_index_end)
 319         {
 320
 321             jnrA             = jjnr[jidx];
 322             j_coord_offsetA  = DIM*jnrA;
 323
 324             /* load j atom coordinates */
 325             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 326                                               &jx0,&jy0,&jz0);
 327
 328             /* Calculate displacement vector */
 329             dx10             = _mm_sub_pd(ix1,jx0);
 330             dy10             = _mm_sub_pd(iy1,jy0);
 331             dz10             = _mm_sub_pd(iz1,jz0);
 332             dx20             = _mm_sub_pd(ix2,jx0);
 333             dy20             = _mm_sub_pd(iy2,jy0);
 334             dz20             = _mm_sub_pd(iz2,jz0);
 335             dx30             = _mm_sub_pd(ix3,jx0);
 336             dy30             = _mm_sub_pd(iy3,jy0);
 337             dz30             = _mm_sub_pd(iz3,jz0);
 338
 339             /* Calculate squared distance and things based on it */
 340             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 341             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 342             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 343
 344             rinv10           = avx128fma_invsqrt_d(rsq10);
 345             rinv20           = avx128fma_invsqrt_d(rsq20);
 346             rinv30           = avx128fma_invsqrt_d(rsq30);
 347
 348             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 349             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 350             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 351
 352             /* Load parameters for j particles */
 353             jq0              = _mm_load_sd(charge+jnrA+0);
 354
 355             fjx0             = _mm_setzero_pd();
 356             fjy0             = _mm_setzero_pd();
 357             fjz0             = _mm_setzero_pd();
 358
 359             /**************************
 360              * CALCULATE INTERACTIONS *
 361              **************************/
 362
 363             if (gmx_mm_any_lt(rsq10,rcutoff2))
 364             {
 365
 366             /* Compute parameters for interactions between i and j atoms */
 367             qq10             = _mm_mul_pd(iq1,jq0);
 368
 369             /* REACTION-FIELD ELECTROSTATICS */
 370             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 371             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 372
 373             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 374
 375             /* Update potential sum for this i atom from the interaction with this j atom. */
 376             velec            = _mm_and_pd(velec,cutoff_mask);
 377             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 378             velecsum         = _mm_add_pd(velecsum,velec);
 379
 380             fscal            = felec;
 381
 382             fscal            = _mm_and_pd(fscal,cutoff_mask);
 383
 384             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 385
 386             /* Update vectorial force */
 387             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 388             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 389             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 390
 391             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 392             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 393             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 394
 395             }
 396
 397             /**************************
 398              * CALCULATE INTERACTIONS *
 399              **************************/
 400
 401             if (gmx_mm_any_lt(rsq20,rcutoff2))
 402             {
 403
 404             /* Compute parameters for interactions between i and j atoms */
 405             qq20             = _mm_mul_pd(iq2,jq0);
 406
 407             /* REACTION-FIELD ELECTROSTATICS */
 408             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 409             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 410
 411             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 412
 413             /* Update potential sum for this i atom from the interaction with this j atom. */
 414             velec            = _mm_and_pd(velec,cutoff_mask);
 415             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 416             velecsum         = _mm_add_pd(velecsum,velec);
 417
 418             fscal            = felec;
 419
 420             fscal            = _mm_and_pd(fscal,cutoff_mask);
 421
 422             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 423
 424             /* Update vectorial force */
 425             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 426             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 427             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 428
 429             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 430             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 431             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 432
 433             }
 434
 435             /**************************
 436              * CALCULATE INTERACTIONS *
 437              **************************/
 438
 439             if (gmx_mm_any_lt(rsq30,rcutoff2))
 440             {
 441
 442             /* Compute parameters for interactions between i and j atoms */
 443             qq30             = _mm_mul_pd(iq3,jq0);
 444
 445             /* REACTION-FIELD ELECTROSTATICS */
 446             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_macc_pd(krf,rsq30,rinv30),crf));
 447             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 448
 449             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 450
 451             /* Update potential sum for this i atom from the interaction with this j atom. */
 452             velec            = _mm_and_pd(velec,cutoff_mask);
 453             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 454             velecsum         = _mm_add_pd(velecsum,velec);
 455
 456             fscal            = felec;
 457
 458             fscal            = _mm_and_pd(fscal,cutoff_mask);
 459
 460             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 461
 462             /* Update vectorial force */
 463             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 464             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 465             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 466
 467             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 468             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 469             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 470
 471             }
 472
 473             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 474
 475             /* Inner loop uses 120 flops */
 476         }
 477
 478         /* End of innermost loop */
 479
 480         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 481                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 482
 483         ggid                        = gid[iidx];
 484         /* Update potential energies */
 485         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 486
 487         /* Increment number of inner iterations */
 488         inneriter                  += j_index_end - j_index_start;
 489
 490         /* Outer loop uses 19 flops */
 491     }
 492
 493     /* Increment number of outer iterations */
 494     outeriter        += nri;
 495
 496     /* Update outer/inner flops */
 497
 498     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*120);
 499 }
 500 /*
 501  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomW4P1_F_avx_128_fma_double
 502  * Electrostatics interaction: ReactionField
 503  * VdW interaction:            None
 504  * Geometry:                   Water4-Particle
 505  * Calculate force/pot:        Force
 506  */
 507 void
 508 nb_kernel_ElecRFCut_VdwNone_GeomW4P1_F_avx_128_fma_double
 509                     (t_nblist                    * gmx_restrict       nlist,
 510                      rvec                        * gmx_restrict          xx,
 511                      rvec                        * gmx_restrict          ff,
 512                      struct t_forcerec           * gmx_restrict          fr,
 513                      t_mdatoms                   * gmx_restrict     mdatoms,
 514                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 515                      t_nrnb                      * gmx_restrict        nrnb)
 516 {
 517     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 518      * just 0 for non-waters.
 519      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 520      * jnr indices corresponding to data put in the four positions in the SIMD register.
 521      */
 522     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 523     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 524     int              jnrA,jnrB;
 525     int              j_coord_offsetA,j_coord_offsetB;
 526     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 527     real             rcutoff_scalar;
 528     real             *shiftvec,*fshift,*x,*f;
 529     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 530     int              vdwioffset1;
 531     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 532     int              vdwioffset2;
 533     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 534     int              vdwioffset3;
 535     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 536     int              vdwjidx0A,vdwjidx0B;
 537     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 538     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 539     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 540     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 541     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 542     real             *charge;
 543     __m128d          dummy_mask,cutoff_mask;
 544     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 545     __m128d          one     = _mm_set1_pd(1.0);
 546     __m128d          two     = _mm_set1_pd(2.0);
 547     x                = xx[0];
 548     f                = ff[0];
 549
 550     nri              = nlist->nri;
 551     iinr             = nlist->iinr;
 552     jindex           = nlist->jindex;
 553     jjnr             = nlist->jjnr;
 554     shiftidx         = nlist->shift;
 555     gid              = nlist->gid;
 556     shiftvec         = fr->shift_vec[0];
 557     fshift           = fr->fshift[0];
 558     facel            = _mm_set1_pd(fr->ic->epsfac);
 559     charge           = mdatoms->chargeA;
 560     krf              = _mm_set1_pd(fr->ic->k_rf);
 561     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 562     crf              = _mm_set1_pd(fr->ic->c_rf);
 563
 564     /* Setup water-specific parameters */
 565     inr              = nlist->iinr[0];
 566     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 567     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 568     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 569
 570     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 571     rcutoff_scalar   = fr->ic->rcoulomb;
 572     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 573     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 574
 575     /* Avoid stupid compiler warnings */
 576     jnrA = jnrB = 0;
 577     j_coord_offsetA = 0;
 578     j_coord_offsetB = 0;
 579
 580     outeriter        = 0;
 581     inneriter        = 0;
 582
 583     /* Start outer loop over neighborlists */
 584     for(iidx=0; iidx<nri; iidx++)
 585     {
 586         /* Load shift vector for this list */
 587         i_shift_offset   = DIM*shiftidx[iidx];
 588
 589         /* Load limits for loop over neighbors */
 590         j_index_start    = jindex[iidx];
 591         j_index_end      = jindex[iidx+1];
 592
 593         /* Get outer coordinate index */
 594         inr              = iinr[iidx];
 595         i_coord_offset   = DIM*inr;
 596
 597         /* Load i particle coords and add shift vector */
 598         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 599                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 600
 601         fix1             = _mm_setzero_pd();
 602         fiy1             = _mm_setzero_pd();
 603         fiz1             = _mm_setzero_pd();
 604         fix2             = _mm_setzero_pd();
 605         fiy2             = _mm_setzero_pd();
 606         fiz2             = _mm_setzero_pd();
 607         fix3             = _mm_setzero_pd();
 608         fiy3             = _mm_setzero_pd();
 609         fiz3             = _mm_setzero_pd();
 610
 611         /* Start inner kernel loop */
 612         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 613         {
 614
 615             /* Get j neighbor index, and coordinate index */
 616             jnrA             = jjnr[jidx];
 617             jnrB             = jjnr[jidx+1];
 618             j_coord_offsetA  = DIM*jnrA;
 619             j_coord_offsetB  = DIM*jnrB;
 620
 621             /* load j atom coordinates */
 622             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 623                                               &jx0,&jy0,&jz0);
 624
 625             /* Calculate displacement vector */
 626             dx10             = _mm_sub_pd(ix1,jx0);
 627             dy10             = _mm_sub_pd(iy1,jy0);
 628             dz10             = _mm_sub_pd(iz1,jz0);
 629             dx20             = _mm_sub_pd(ix2,jx0);
 630             dy20             = _mm_sub_pd(iy2,jy0);
 631             dz20             = _mm_sub_pd(iz2,jz0);
 632             dx30             = _mm_sub_pd(ix3,jx0);
 633             dy30             = _mm_sub_pd(iy3,jy0);
 634             dz30             = _mm_sub_pd(iz3,jz0);
 635
 636             /* Calculate squared distance and things based on it */
 637             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 638             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 639             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 640
 641             rinv10           = avx128fma_invsqrt_d(rsq10);
 642             rinv20           = avx128fma_invsqrt_d(rsq20);
 643             rinv30           = avx128fma_invsqrt_d(rsq30);
 644
 645             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 646             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 647             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 648
 649             /* Load parameters for j particles */
 650             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 651
 652             fjx0             = _mm_setzero_pd();
 653             fjy0             = _mm_setzero_pd();
 654             fjz0             = _mm_setzero_pd();
 655
 656             /**************************
 657              * CALCULATE INTERACTIONS *
 658              **************************/
 659
 660             if (gmx_mm_any_lt(rsq10,rcutoff2))
 661             {
 662
 663             /* Compute parameters for interactions between i and j atoms */
 664             qq10             = _mm_mul_pd(iq1,jq0);
 665
 666             /* REACTION-FIELD ELECTROSTATICS */
 667             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 668
 669             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 670
 671             fscal            = felec;
 672
 673             fscal            = _mm_and_pd(fscal,cutoff_mask);
 674
 675             /* Update vectorial force */
 676             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 677             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 678             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 679
 680             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 681             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 682             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 683
 684             }
 685
 686             /**************************
 687              * CALCULATE INTERACTIONS *
 688              **************************/
 689
 690             if (gmx_mm_any_lt(rsq20,rcutoff2))
 691             {
 692
 693             /* Compute parameters for interactions between i and j atoms */
 694             qq20             = _mm_mul_pd(iq2,jq0);
 695
 696             /* REACTION-FIELD ELECTROSTATICS */
 697             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 698
 699             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 700
 701             fscal            = felec;
 702
 703             fscal            = _mm_and_pd(fscal,cutoff_mask);
 704
 705             /* Update vectorial force */
 706             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 707             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 708             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 709
 710             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 711             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 712             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 713
 714             }
 715
 716             /**************************
 717              * CALCULATE INTERACTIONS *
 718              **************************/
 719
 720             if (gmx_mm_any_lt(rsq30,rcutoff2))
 721             {
 722
 723             /* Compute parameters for interactions between i and j atoms */
 724             qq30             = _mm_mul_pd(iq3,jq0);
 725
 726             /* REACTION-FIELD ELECTROSTATICS */
 727             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 728
 729             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 730
 731             fscal            = felec;
 732
 733             fscal            = _mm_and_pd(fscal,cutoff_mask);
 734
 735             /* Update vectorial force */
 736             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 737             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 738             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 739
 740             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 741             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 742             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 743
 744             }
 745
 746             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 747
 748             /* Inner loop uses 102 flops */
 749         }
 750
 751         if(jidx<j_index_end)
 752         {
 753
 754             jnrA             = jjnr[jidx];
 755             j_coord_offsetA  = DIM*jnrA;
 756
 757             /* load j atom coordinates */
 758             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 759                                               &jx0,&jy0,&jz0);
 760
 761             /* Calculate displacement vector */
 762             dx10             = _mm_sub_pd(ix1,jx0);
 763             dy10             = _mm_sub_pd(iy1,jy0);
 764             dz10             = _mm_sub_pd(iz1,jz0);
 765             dx20             = _mm_sub_pd(ix2,jx0);
 766             dy20             = _mm_sub_pd(iy2,jy0);
 767             dz20             = _mm_sub_pd(iz2,jz0);
 768             dx30             = _mm_sub_pd(ix3,jx0);
 769             dy30             = _mm_sub_pd(iy3,jy0);
 770             dz30             = _mm_sub_pd(iz3,jz0);
 771
 772             /* Calculate squared distance and things based on it */
 773             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 774             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 775             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 776
 777             rinv10           = avx128fma_invsqrt_d(rsq10);
 778             rinv20           = avx128fma_invsqrt_d(rsq20);
 779             rinv30           = avx128fma_invsqrt_d(rsq30);
 780
 781             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 782             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 783             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 784
 785             /* Load parameters for j particles */
 786             jq0              = _mm_load_sd(charge+jnrA+0);
 787
 788             fjx0             = _mm_setzero_pd();
 789             fjy0             = _mm_setzero_pd();
 790             fjz0             = _mm_setzero_pd();
 791
 792             /**************************
 793              * CALCULATE INTERACTIONS *
 794              **************************/
 795
 796             if (gmx_mm_any_lt(rsq10,rcutoff2))
 797             {
 798
 799             /* Compute parameters for interactions between i and j atoms */
 800             qq10             = _mm_mul_pd(iq1,jq0);
 801
 802             /* REACTION-FIELD ELECTROSTATICS */
 803             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 804
 805             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 806
 807             fscal            = felec;
 808
 809             fscal            = _mm_and_pd(fscal,cutoff_mask);
 810
 811             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 812
 813             /* Update vectorial force */
 814             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 815             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 816             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 817
 818             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 819             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 820             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 821
 822             }
 823
 824             /**************************
 825              * CALCULATE INTERACTIONS *
 826              **************************/
 827
 828             if (gmx_mm_any_lt(rsq20,rcutoff2))
 829             {
 830
 831             /* Compute parameters for interactions between i and j atoms */
 832             qq20             = _mm_mul_pd(iq2,jq0);
 833
 834             /* REACTION-FIELD ELECTROSTATICS */
 835             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 836
 837             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 838
 839             fscal            = felec;
 840
 841             fscal            = _mm_and_pd(fscal,cutoff_mask);
 842
 843             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 844
 845             /* Update vectorial force */
 846             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 847             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 848             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 849
 850             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 851             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 852             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 853
 854             }
 855
 856             /**************************
 857              * CALCULATE INTERACTIONS *
 858              **************************/
 859
 860             if (gmx_mm_any_lt(rsq30,rcutoff2))
 861             {
 862
 863             /* Compute parameters for interactions between i and j atoms */
 864             qq30             = _mm_mul_pd(iq3,jq0);
 865
 866             /* REACTION-FIELD ELECTROSTATICS */
 867             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 868
 869             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 870
 871             fscal            = felec;
 872
 873             fscal            = _mm_and_pd(fscal,cutoff_mask);
 874
 875             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 876
 877             /* Update vectorial force */
 878             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 879             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 880             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 881
 882             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 883             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 884             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 885
 886             }
 887
 888             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 889
 890             /* Inner loop uses 102 flops */
 891         }
 892
 893         /* End of innermost loop */
 894
 895         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 896                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 897
 898         /* Increment number of inner iterations */
 899         inneriter                  += j_index_end - j_index_start;
 900
 901         /* Outer loop uses 18 flops */
 902     }
 903
 904     /* Increment number of outer iterations */
 905     outeriter        += nri;
 906
 907     /* Update outer/inner flops */
 908
 909     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*102);
 910 }