src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecRFCut_VdwNone_GeomW3P1_sse4_1_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 sse4_1_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_sse4_1_double.h"
  50 #include "kernelutil_x86_sse4_1_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomW3P1_VF_sse4_1_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            None
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwNone_GeomW3P1_VF_sse4_1_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              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  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     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 119     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 120     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 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,
 151                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 152
 153         fix0             = _mm_setzero_pd();
 154         fiy0             = _mm_setzero_pd();
 155         fiz0             = _mm_setzero_pd();
 156         fix1             = _mm_setzero_pd();
 157         fiy1             = _mm_setzero_pd();
 158         fiz1             = _mm_setzero_pd();
 159         fix2             = _mm_setzero_pd();
 160         fiy2             = _mm_setzero_pd();
 161         fiz2             = _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             dx00             = _mm_sub_pd(ix0,jx0);
 182             dy00             = _mm_sub_pd(iy0,jy0);
 183             dz00             = _mm_sub_pd(iz0,jz0);
 184             dx10             = _mm_sub_pd(ix1,jx0);
 185             dy10             = _mm_sub_pd(iy1,jy0);
 186             dz10             = _mm_sub_pd(iz1,jz0);
 187             dx20             = _mm_sub_pd(ix2,jx0);
 188             dy20             = _mm_sub_pd(iy2,jy0);
 189             dz20             = _mm_sub_pd(iz2,jz0);
 190
 191             /* Calculate squared distance and things based on it */
 192             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 193             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 194             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 195
 196             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 197             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 198             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 199
 200             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 201             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 202             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 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(rsq00,rcutoff2))
 216             {
 217
 218             /* Compute parameters for interactions between i and j atoms */
 219             qq00             = _mm_mul_pd(iq0,jq0);
 220
 221             /* REACTION-FIELD ELECTROSTATICS */
 222             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 223             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 224
 225             cutoff_mask      = _mm_cmplt_pd(rsq00,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             /* Calculate temporary vectorial force */
 236             tx               = _mm_mul_pd(fscal,dx00);
 237             ty               = _mm_mul_pd(fscal,dy00);
 238             tz               = _mm_mul_pd(fscal,dz00);
 239
 240             /* Update vectorial force */
 241             fix0             = _mm_add_pd(fix0,tx);
 242             fiy0             = _mm_add_pd(fiy0,ty);
 243             fiz0             = _mm_add_pd(fiz0,tz);
 244
 245             fjx0             = _mm_add_pd(fjx0,tx);
 246             fjy0             = _mm_add_pd(fjy0,ty);
 247             fjz0             = _mm_add_pd(fjz0,tz);
 248
 249             }
 250
 251             /**************************
 252              * CALCULATE INTERACTIONS *
 253              **************************/
 254
 255             if (gmx_mm_any_lt(rsq10,rcutoff2))
 256             {
 257
 258             /* Compute parameters for interactions between i and j atoms */
 259             qq10             = _mm_mul_pd(iq1,jq0);
 260
 261             /* REACTION-FIELD ELECTROSTATICS */
 262             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 263             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 264
 265             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 266
 267             /* Update potential sum for this i atom from the interaction with this j atom. */
 268             velec            = _mm_and_pd(velec,cutoff_mask);
 269             velecsum         = _mm_add_pd(velecsum,velec);
 270
 271             fscal            = felec;
 272
 273             fscal            = _mm_and_pd(fscal,cutoff_mask);
 274
 275             /* Calculate temporary vectorial force */
 276             tx               = _mm_mul_pd(fscal,dx10);
 277             ty               = _mm_mul_pd(fscal,dy10);
 278             tz               = _mm_mul_pd(fscal,dz10);
 279
 280             /* Update vectorial force */
 281             fix1             = _mm_add_pd(fix1,tx);
 282             fiy1             = _mm_add_pd(fiy1,ty);
 283             fiz1             = _mm_add_pd(fiz1,tz);
 284
 285             fjx0             = _mm_add_pd(fjx0,tx);
 286             fjy0             = _mm_add_pd(fjy0,ty);
 287             fjz0             = _mm_add_pd(fjz0,tz);
 288
 289             }
 290
 291             /**************************
 292              * CALCULATE INTERACTIONS *
 293              **************************/
 294
 295             if (gmx_mm_any_lt(rsq20,rcutoff2))
 296             {
 297
 298             /* Compute parameters for interactions between i and j atoms */
 299             qq20             = _mm_mul_pd(iq2,jq0);
 300
 301             /* REACTION-FIELD ELECTROSTATICS */
 302             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 303             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 304
 305             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 306
 307             /* Update potential sum for this i atom from the interaction with this j atom. */
 308             velec            = _mm_and_pd(velec,cutoff_mask);
 309             velecsum         = _mm_add_pd(velecsum,velec);
 310
 311             fscal            = felec;
 312
 313             fscal            = _mm_and_pd(fscal,cutoff_mask);
 314
 315             /* Calculate temporary vectorial force */
 316             tx               = _mm_mul_pd(fscal,dx20);
 317             ty               = _mm_mul_pd(fscal,dy20);
 318             tz               = _mm_mul_pd(fscal,dz20);
 319
 320             /* Update vectorial force */
 321             fix2             = _mm_add_pd(fix2,tx);
 322             fiy2             = _mm_add_pd(fiy2,ty);
 323             fiz2             = _mm_add_pd(fiz2,tz);
 324
 325             fjx0             = _mm_add_pd(fjx0,tx);
 326             fjy0             = _mm_add_pd(fjy0,ty);
 327             fjz0             = _mm_add_pd(fjz0,tz);
 328
 329             }
 330
 331             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 332
 333             /* Inner loop uses 111 flops */
 334         }
 335
 336         if(jidx<j_index_end)
 337         {
 338
 339             jnrA             = jjnr[jidx];
 340             j_coord_offsetA  = DIM*jnrA;
 341
 342             /* load j atom coordinates */
 343             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 344                                               &jx0,&jy0,&jz0);
 345
 346             /* Calculate displacement vector */
 347             dx00             = _mm_sub_pd(ix0,jx0);
 348             dy00             = _mm_sub_pd(iy0,jy0);
 349             dz00             = _mm_sub_pd(iz0,jz0);
 350             dx10             = _mm_sub_pd(ix1,jx0);
 351             dy10             = _mm_sub_pd(iy1,jy0);
 352             dz10             = _mm_sub_pd(iz1,jz0);
 353             dx20             = _mm_sub_pd(ix2,jx0);
 354             dy20             = _mm_sub_pd(iy2,jy0);
 355             dz20             = _mm_sub_pd(iz2,jz0);
 356
 357             /* Calculate squared distance and things based on it */
 358             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 359             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 360             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 361
 362             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 363             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 364             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 365
 366             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 367             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 368             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 369
 370             /* Load parameters for j particles */
 371             jq0              = _mm_load_sd(charge+jnrA+0);
 372
 373             fjx0             = _mm_setzero_pd();
 374             fjy0             = _mm_setzero_pd();
 375             fjz0             = _mm_setzero_pd();
 376
 377             /**************************
 378              * CALCULATE INTERACTIONS *
 379              **************************/
 380
 381             if (gmx_mm_any_lt(rsq00,rcutoff2))
 382             {
 383
 384             /* Compute parameters for interactions between i and j atoms */
 385             qq00             = _mm_mul_pd(iq0,jq0);
 386
 387             /* REACTION-FIELD ELECTROSTATICS */
 388             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 389             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 390
 391             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 392
 393             /* Update potential sum for this i atom from the interaction with this j atom. */
 394             velec            = _mm_and_pd(velec,cutoff_mask);
 395             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 396             velecsum         = _mm_add_pd(velecsum,velec);
 397
 398             fscal            = felec;
 399
 400             fscal            = _mm_and_pd(fscal,cutoff_mask);
 401
 402             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 403
 404             /* Calculate temporary vectorial force */
 405             tx               = _mm_mul_pd(fscal,dx00);
 406             ty               = _mm_mul_pd(fscal,dy00);
 407             tz               = _mm_mul_pd(fscal,dz00);
 408
 409             /* Update vectorial force */
 410             fix0             = _mm_add_pd(fix0,tx);
 411             fiy0             = _mm_add_pd(fiy0,ty);
 412             fiz0             = _mm_add_pd(fiz0,tz);
 413
 414             fjx0             = _mm_add_pd(fjx0,tx);
 415             fjy0             = _mm_add_pd(fjy0,ty);
 416             fjz0             = _mm_add_pd(fjz0,tz);
 417
 418             }
 419
 420             /**************************
 421              * CALCULATE INTERACTIONS *
 422              **************************/
 423
 424             if (gmx_mm_any_lt(rsq10,rcutoff2))
 425             {
 426
 427             /* Compute parameters for interactions between i and j atoms */
 428             qq10             = _mm_mul_pd(iq1,jq0);
 429
 430             /* REACTION-FIELD ELECTROSTATICS */
 431             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 432             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 433
 434             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 435
 436             /* Update potential sum for this i atom from the interaction with this j atom. */
 437             velec            = _mm_and_pd(velec,cutoff_mask);
 438             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 439             velecsum         = _mm_add_pd(velecsum,velec);
 440
 441             fscal            = felec;
 442
 443             fscal            = _mm_and_pd(fscal,cutoff_mask);
 444
 445             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 446
 447             /* Calculate temporary vectorial force */
 448             tx               = _mm_mul_pd(fscal,dx10);
 449             ty               = _mm_mul_pd(fscal,dy10);
 450             tz               = _mm_mul_pd(fscal,dz10);
 451
 452             /* Update vectorial force */
 453             fix1             = _mm_add_pd(fix1,tx);
 454             fiy1             = _mm_add_pd(fiy1,ty);
 455             fiz1             = _mm_add_pd(fiz1,tz);
 456
 457             fjx0             = _mm_add_pd(fjx0,tx);
 458             fjy0             = _mm_add_pd(fjy0,ty);
 459             fjz0             = _mm_add_pd(fjz0,tz);
 460
 461             }
 462
 463             /**************************
 464              * CALCULATE INTERACTIONS *
 465              **************************/
 466
 467             if (gmx_mm_any_lt(rsq20,rcutoff2))
 468             {
 469
 470             /* Compute parameters for interactions between i and j atoms */
 471             qq20             = _mm_mul_pd(iq2,jq0);
 472
 473             /* REACTION-FIELD ELECTROSTATICS */
 474             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 475             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 476
 477             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 478
 479             /* Update potential sum for this i atom from the interaction with this j atom. */
 480             velec            = _mm_and_pd(velec,cutoff_mask);
 481             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 482             velecsum         = _mm_add_pd(velecsum,velec);
 483
 484             fscal            = felec;
 485
 486             fscal            = _mm_and_pd(fscal,cutoff_mask);
 487
 488             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 489
 490             /* Calculate temporary vectorial force */
 491             tx               = _mm_mul_pd(fscal,dx20);
 492             ty               = _mm_mul_pd(fscal,dy20);
 493             tz               = _mm_mul_pd(fscal,dz20);
 494
 495             /* Update vectorial force */
 496             fix2             = _mm_add_pd(fix2,tx);
 497             fiy2             = _mm_add_pd(fiy2,ty);
 498             fiz2             = _mm_add_pd(fiz2,tz);
 499
 500             fjx0             = _mm_add_pd(fjx0,tx);
 501             fjy0             = _mm_add_pd(fjy0,ty);
 502             fjz0             = _mm_add_pd(fjz0,tz);
 503
 504             }
 505
 506             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 507
 508             /* Inner loop uses 111 flops */
 509         }
 510
 511         /* End of innermost loop */
 512
 513         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 514                                               f+i_coord_offset,fshift+i_shift_offset);
 515
 516         ggid                        = gid[iidx];
 517         /* Update potential energies */
 518         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 519
 520         /* Increment number of inner iterations */
 521         inneriter                  += j_index_end - j_index_start;
 522
 523         /* Outer loop uses 19 flops */
 524     }
 525
 526     /* Increment number of outer iterations */
 527     outeriter        += nri;
 528
 529     /* Update outer/inner flops */
 530
 531     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*111);
 532 }
 533 /*
 534  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomW3P1_F_sse4_1_double
 535  * Electrostatics interaction: ReactionField
 536  * VdW interaction:            None
 537  * Geometry:                   Water3-Particle
 538  * Calculate force/pot:        Force
 539  */
 540 void
 541 nb_kernel_ElecRFCut_VdwNone_GeomW3P1_F_sse4_1_double
 542                     (t_nblist                    * gmx_restrict       nlist,
 543                      rvec                        * gmx_restrict          xx,
 544                      rvec                        * gmx_restrict          ff,
 545                      t_forcerec                  * gmx_restrict          fr,
 546                      t_mdatoms                   * gmx_restrict     mdatoms,
 547                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 548                      t_nrnb                      * gmx_restrict        nrnb)
 549 {
 550     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 551      * just 0 for non-waters.
 552      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 553      * jnr indices corresponding to data put in the four positions in the SIMD register.
 554      */
 555     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 556     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 557     int              jnrA,jnrB;
 558     int              j_coord_offsetA,j_coord_offsetB;
 559     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 560     real             rcutoff_scalar;
 561     real             *shiftvec,*fshift,*x,*f;
 562     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 563     int              vdwioffset0;
 564     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 565     int              vdwioffset1;
 566     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 567     int              vdwioffset2;
 568     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 569     int              vdwjidx0A,vdwjidx0B;
 570     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 571     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 572     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 573     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 574     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 575     real             *charge;
 576     __m128d          dummy_mask,cutoff_mask;
 577     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 578     __m128d          one     = _mm_set1_pd(1.0);
 579     __m128d          two     = _mm_set1_pd(2.0);
 580     x                = xx[0];
 581     f                = ff[0];
 582
 583     nri              = nlist->nri;
 584     iinr             = nlist->iinr;
 585     jindex           = nlist->jindex;
 586     jjnr             = nlist->jjnr;
 587     shiftidx         = nlist->shift;
 588     gid              = nlist->gid;
 589     shiftvec         = fr->shift_vec[0];
 590     fshift           = fr->fshift[0];
 591     facel            = _mm_set1_pd(fr->epsfac);
 592     charge           = mdatoms->chargeA;
 593     krf              = _mm_set1_pd(fr->ic->k_rf);
 594     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 595     crf              = _mm_set1_pd(fr->ic->c_rf);
 596
 597     /* Setup water-specific parameters */
 598     inr              = nlist->iinr[0];
 599     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 600     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 601     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 602
 603     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 604     rcutoff_scalar   = fr->rcoulomb;
 605     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 606     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 607
 608     /* Avoid stupid compiler warnings */
 609     jnrA = jnrB = 0;
 610     j_coord_offsetA = 0;
 611     j_coord_offsetB = 0;
 612
 613     outeriter        = 0;
 614     inneriter        = 0;
 615
 616     /* Start outer loop over neighborlists */
 617     for(iidx=0; iidx<nri; iidx++)
 618     {
 619         /* Load shift vector for this list */
 620         i_shift_offset   = DIM*shiftidx[iidx];
 621
 622         /* Load limits for loop over neighbors */
 623         j_index_start    = jindex[iidx];
 624         j_index_end      = jindex[iidx+1];
 625
 626         /* Get outer coordinate index */
 627         inr              = iinr[iidx];
 628         i_coord_offset   = DIM*inr;
 629
 630         /* Load i particle coords and add shift vector */
 631         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 632                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 633
 634         fix0             = _mm_setzero_pd();
 635         fiy0             = _mm_setzero_pd();
 636         fiz0             = _mm_setzero_pd();
 637         fix1             = _mm_setzero_pd();
 638         fiy1             = _mm_setzero_pd();
 639         fiz1             = _mm_setzero_pd();
 640         fix2             = _mm_setzero_pd();
 641         fiy2             = _mm_setzero_pd();
 642         fiz2             = _mm_setzero_pd();
 643
 644         /* Start inner kernel loop */
 645         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 646         {
 647
 648             /* Get j neighbor index, and coordinate index */
 649             jnrA             = jjnr[jidx];
 650             jnrB             = jjnr[jidx+1];
 651             j_coord_offsetA  = DIM*jnrA;
 652             j_coord_offsetB  = DIM*jnrB;
 653
 654             /* load j atom coordinates */
 655             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 656                                               &jx0,&jy0,&jz0);
 657
 658             /* Calculate displacement vector */
 659             dx00             = _mm_sub_pd(ix0,jx0);
 660             dy00             = _mm_sub_pd(iy0,jy0);
 661             dz00             = _mm_sub_pd(iz0,jz0);
 662             dx10             = _mm_sub_pd(ix1,jx0);
 663             dy10             = _mm_sub_pd(iy1,jy0);
 664             dz10             = _mm_sub_pd(iz1,jz0);
 665             dx20             = _mm_sub_pd(ix2,jx0);
 666             dy20             = _mm_sub_pd(iy2,jy0);
 667             dz20             = _mm_sub_pd(iz2,jz0);
 668
 669             /* Calculate squared distance and things based on it */
 670             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 671             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 672             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 673
 674             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 675             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 676             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 677
 678             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 679             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 680             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 681
 682             /* Load parameters for j particles */
 683             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 684
 685             fjx0             = _mm_setzero_pd();
 686             fjy0             = _mm_setzero_pd();
 687             fjz0             = _mm_setzero_pd();
 688
 689             /**************************
 690              * CALCULATE INTERACTIONS *
 691              **************************/
 692
 693             if (gmx_mm_any_lt(rsq00,rcutoff2))
 694             {
 695
 696             /* Compute parameters for interactions between i and j atoms */
 697             qq00             = _mm_mul_pd(iq0,jq0);
 698
 699             /* REACTION-FIELD ELECTROSTATICS */
 700             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 701
 702             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 703
 704             fscal            = felec;
 705
 706             fscal            = _mm_and_pd(fscal,cutoff_mask);
 707
 708             /* Calculate temporary vectorial force */
 709             tx               = _mm_mul_pd(fscal,dx00);
 710             ty               = _mm_mul_pd(fscal,dy00);
 711             tz               = _mm_mul_pd(fscal,dz00);
 712
 713             /* Update vectorial force */
 714             fix0             = _mm_add_pd(fix0,tx);
 715             fiy0             = _mm_add_pd(fiy0,ty);
 716             fiz0             = _mm_add_pd(fiz0,tz);
 717
 718             fjx0             = _mm_add_pd(fjx0,tx);
 719             fjy0             = _mm_add_pd(fjy0,ty);
 720             fjz0             = _mm_add_pd(fjz0,tz);
 721
 722             }
 723
 724             /**************************
 725              * CALCULATE INTERACTIONS *
 726              **************************/
 727
 728             if (gmx_mm_any_lt(rsq10,rcutoff2))
 729             {
 730
 731             /* Compute parameters for interactions between i and j atoms */
 732             qq10             = _mm_mul_pd(iq1,jq0);
 733
 734             /* REACTION-FIELD ELECTROSTATICS */
 735             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 736
 737             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 738
 739             fscal            = felec;
 740
 741             fscal            = _mm_and_pd(fscal,cutoff_mask);
 742
 743             /* Calculate temporary vectorial force */
 744             tx               = _mm_mul_pd(fscal,dx10);
 745             ty               = _mm_mul_pd(fscal,dy10);
 746             tz               = _mm_mul_pd(fscal,dz10);
 747
 748             /* Update vectorial force */
 749             fix1             = _mm_add_pd(fix1,tx);
 750             fiy1             = _mm_add_pd(fiy1,ty);
 751             fiz1             = _mm_add_pd(fiz1,tz);
 752
 753             fjx0             = _mm_add_pd(fjx0,tx);
 754             fjy0             = _mm_add_pd(fjy0,ty);
 755             fjz0             = _mm_add_pd(fjz0,tz);
 756
 757             }
 758
 759             /**************************
 760              * CALCULATE INTERACTIONS *
 761              **************************/
 762
 763             if (gmx_mm_any_lt(rsq20,rcutoff2))
 764             {
 765
 766             /* Compute parameters for interactions between i and j atoms */
 767             qq20             = _mm_mul_pd(iq2,jq0);
 768
 769             /* REACTION-FIELD ELECTROSTATICS */
 770             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 771
 772             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 773
 774             fscal            = felec;
 775
 776             fscal            = _mm_and_pd(fscal,cutoff_mask);
 777
 778             /* Calculate temporary vectorial force */
 779             tx               = _mm_mul_pd(fscal,dx20);
 780             ty               = _mm_mul_pd(fscal,dy20);
 781             tz               = _mm_mul_pd(fscal,dz20);
 782
 783             /* Update vectorial force */
 784             fix2             = _mm_add_pd(fix2,tx);
 785             fiy2             = _mm_add_pd(fiy2,ty);
 786             fiz2             = _mm_add_pd(fiz2,tz);
 787
 788             fjx0             = _mm_add_pd(fjx0,tx);
 789             fjy0             = _mm_add_pd(fjy0,ty);
 790             fjz0             = _mm_add_pd(fjz0,tz);
 791
 792             }
 793
 794             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 795
 796             /* Inner loop uses 93 flops */
 797         }
 798
 799         if(jidx<j_index_end)
 800         {
 801
 802             jnrA             = jjnr[jidx];
 803             j_coord_offsetA  = DIM*jnrA;
 804
 805             /* load j atom coordinates */
 806             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 807                                               &jx0,&jy0,&jz0);
 808
 809             /* Calculate displacement vector */
 810             dx00             = _mm_sub_pd(ix0,jx0);
 811             dy00             = _mm_sub_pd(iy0,jy0);
 812             dz00             = _mm_sub_pd(iz0,jz0);
 813             dx10             = _mm_sub_pd(ix1,jx0);
 814             dy10             = _mm_sub_pd(iy1,jy0);
 815             dz10             = _mm_sub_pd(iz1,jz0);
 816             dx20             = _mm_sub_pd(ix2,jx0);
 817             dy20             = _mm_sub_pd(iy2,jy0);
 818             dz20             = _mm_sub_pd(iz2,jz0);
 819
 820             /* Calculate squared distance and things based on it */
 821             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 822             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 823             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 824
 825             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 826             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 827             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 828
 829             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 830             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 831             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 832
 833             /* Load parameters for j particles */
 834             jq0              = _mm_load_sd(charge+jnrA+0);
 835
 836             fjx0             = _mm_setzero_pd();
 837             fjy0             = _mm_setzero_pd();
 838             fjz0             = _mm_setzero_pd();
 839
 840             /**************************
 841              * CALCULATE INTERACTIONS *
 842              **************************/
 843
 844             if (gmx_mm_any_lt(rsq00,rcutoff2))
 845             {
 846
 847             /* Compute parameters for interactions between i and j atoms */
 848             qq00             = _mm_mul_pd(iq0,jq0);
 849
 850             /* REACTION-FIELD ELECTROSTATICS */
 851             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 852
 853             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 854
 855             fscal            = felec;
 856
 857             fscal            = _mm_and_pd(fscal,cutoff_mask);
 858
 859             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 860
 861             /* Calculate temporary vectorial force */
 862             tx               = _mm_mul_pd(fscal,dx00);
 863             ty               = _mm_mul_pd(fscal,dy00);
 864             tz               = _mm_mul_pd(fscal,dz00);
 865
 866             /* Update vectorial force */
 867             fix0             = _mm_add_pd(fix0,tx);
 868             fiy0             = _mm_add_pd(fiy0,ty);
 869             fiz0             = _mm_add_pd(fiz0,tz);
 870
 871             fjx0             = _mm_add_pd(fjx0,tx);
 872             fjy0             = _mm_add_pd(fjy0,ty);
 873             fjz0             = _mm_add_pd(fjz0,tz);
 874
 875             }
 876
 877             /**************************
 878              * CALCULATE INTERACTIONS *
 879              **************************/
 880
 881             if (gmx_mm_any_lt(rsq10,rcutoff2))
 882             {
 883
 884             /* Compute parameters for interactions between i and j atoms */
 885             qq10             = _mm_mul_pd(iq1,jq0);
 886
 887             /* REACTION-FIELD ELECTROSTATICS */
 888             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 889
 890             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 891
 892             fscal            = felec;
 893
 894             fscal            = _mm_and_pd(fscal,cutoff_mask);
 895
 896             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 897
 898             /* Calculate temporary vectorial force */
 899             tx               = _mm_mul_pd(fscal,dx10);
 900             ty               = _mm_mul_pd(fscal,dy10);
 901             tz               = _mm_mul_pd(fscal,dz10);
 902
 903             /* Update vectorial force */
 904             fix1             = _mm_add_pd(fix1,tx);
 905             fiy1             = _mm_add_pd(fiy1,ty);
 906             fiz1             = _mm_add_pd(fiz1,tz);
 907
 908             fjx0             = _mm_add_pd(fjx0,tx);
 909             fjy0             = _mm_add_pd(fjy0,ty);
 910             fjz0             = _mm_add_pd(fjz0,tz);
 911
 912             }
 913
 914             /**************************
 915              * CALCULATE INTERACTIONS *
 916              **************************/
 917
 918             if (gmx_mm_any_lt(rsq20,rcutoff2))
 919             {
 920
 921             /* Compute parameters for interactions between i and j atoms */
 922             qq20             = _mm_mul_pd(iq2,jq0);
 923
 924             /* REACTION-FIELD ELECTROSTATICS */
 925             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 926
 927             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 928
 929             fscal            = felec;
 930
 931             fscal            = _mm_and_pd(fscal,cutoff_mask);
 932
 933             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 934
 935             /* Calculate temporary vectorial force */
 936             tx               = _mm_mul_pd(fscal,dx20);
 937             ty               = _mm_mul_pd(fscal,dy20);
 938             tz               = _mm_mul_pd(fscal,dz20);
 939
 940             /* Update vectorial force */
 941             fix2             = _mm_add_pd(fix2,tx);
 942             fiy2             = _mm_add_pd(fiy2,ty);
 943             fiz2             = _mm_add_pd(fiz2,tz);
 944
 945             fjx0             = _mm_add_pd(fjx0,tx);
 946             fjy0             = _mm_add_pd(fjy0,ty);
 947             fjz0             = _mm_add_pd(fjz0,tz);
 948
 949             }
 950
 951             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 952
 953             /* Inner loop uses 93 flops */
 954         }
 955
 956         /* End of innermost loop */
 957
 958         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 959                                               f+i_coord_offset,fshift+i_shift_offset);
 960
 961         /* Increment number of inner iterations */
 962         inneriter                  += j_index_end - j_index_start;
 963
 964         /* Outer loop uses 18 flops */
 965     }
 966
 967     /* Increment number of outer iterations */
 968     outeriter        += nri;
 969
 970     /* Update outer/inner flops */
 971
 972     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*93);
 973 }