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) 2012,2013,2014,2015,2017, 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 sse4_1_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_sse4_1_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomW3P1_VF_sse4_1_double
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            None
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRFCut_VdwNone_GeomW3P1_VF_sse4_1_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwioffset1;
  82     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  83     int              vdwioffset2;
  84     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  85     int              vdwjidx0A,vdwjidx0B;
  86     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  89     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  90     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     __m128d          dummy_mask,cutoff_mask;
  93     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  94     __m128d          one     = _mm_set1_pd(1.0);
  95     __m128d          two     = _mm_set1_pd(2.0);
  96     x                = xx[0];
  97     f                = ff[0];
  98
  99     nri              = nlist->nri;
 100     iinr             = nlist->iinr;
 101     jindex           = nlist->jindex;
 102     jjnr             = nlist->jjnr;
 103     shiftidx         = nlist->shift;
 104     gid              = nlist->gid;
 105     shiftvec         = fr->shift_vec[0];
 106     fshift           = fr->fshift[0];
 107     facel            = _mm_set1_pd(fr->ic->epsfac);
 108     charge           = mdatoms->chargeA;
 109     krf              = _mm_set1_pd(fr->ic->k_rf);
 110     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 111     crf              = _mm_set1_pd(fr->ic->c_rf);
 112
 113     /* Setup water-specific parameters */
 114     inr              = nlist->iinr[0];
 115     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 116     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 117     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 118
 119     /* 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,
 148                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 149
 150         fix0             = _mm_setzero_pd();
 151         fiy0             = _mm_setzero_pd();
 152         fiz0             = _mm_setzero_pd();
 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
 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             dx00             = _mm_sub_pd(ix0,jx0);
 179             dy00             = _mm_sub_pd(iy0,jy0);
 180             dz00             = _mm_sub_pd(iz0,jz0);
 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
 188             /* Calculate squared distance and things based on it */
 189             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 190             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 191             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 192
 193             rinv00           = sse41_invsqrt_d(rsq00);
 194             rinv10           = sse41_invsqrt_d(rsq10);
 195             rinv20           = sse41_invsqrt_d(rsq20);
 196
 197             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 198             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 199             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 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(rsq00,rcutoff2))
 213             {
 214
 215             /* Compute parameters for interactions between i and j atoms */
 216             qq00             = _mm_mul_pd(iq0,jq0);
 217
 218             /* REACTION-FIELD ELECTROSTATICS */
 219             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 220             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 221
 222             cutoff_mask      = _mm_cmplt_pd(rsq00,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             /* Calculate temporary vectorial force */
 233             tx               = _mm_mul_pd(fscal,dx00);
 234             ty               = _mm_mul_pd(fscal,dy00);
 235             tz               = _mm_mul_pd(fscal,dz00);
 236
 237             /* Update vectorial force */
 238             fix0             = _mm_add_pd(fix0,tx);
 239             fiy0             = _mm_add_pd(fiy0,ty);
 240             fiz0             = _mm_add_pd(fiz0,tz);
 241
 242             fjx0             = _mm_add_pd(fjx0,tx);
 243             fjy0             = _mm_add_pd(fjy0,ty);
 244             fjz0             = _mm_add_pd(fjz0,tz);
 245
 246             }
 247
 248             /**************************
 249              * CALCULATE INTERACTIONS *
 250              **************************/
 251
 252             if (gmx_mm_any_lt(rsq10,rcutoff2))
 253             {
 254
 255             /* Compute parameters for interactions between i and j atoms */
 256             qq10             = _mm_mul_pd(iq1,jq0);
 257
 258             /* REACTION-FIELD ELECTROSTATICS */
 259             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 260             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 261
 262             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 263
 264             /* Update potential sum for this i atom from the interaction with this j atom. */
 265             velec            = _mm_and_pd(velec,cutoff_mask);
 266             velecsum         = _mm_add_pd(velecsum,velec);
 267
 268             fscal            = felec;
 269
 270             fscal            = _mm_and_pd(fscal,cutoff_mask);
 271
 272             /* Calculate temporary vectorial force */
 273             tx               = _mm_mul_pd(fscal,dx10);
 274             ty               = _mm_mul_pd(fscal,dy10);
 275             tz               = _mm_mul_pd(fscal,dz10);
 276
 277             /* Update vectorial force */
 278             fix1             = _mm_add_pd(fix1,tx);
 279             fiy1             = _mm_add_pd(fiy1,ty);
 280             fiz1             = _mm_add_pd(fiz1,tz);
 281
 282             fjx0             = _mm_add_pd(fjx0,tx);
 283             fjy0             = _mm_add_pd(fjy0,ty);
 284             fjz0             = _mm_add_pd(fjz0,tz);
 285
 286             }
 287
 288             /**************************
 289              * CALCULATE INTERACTIONS *
 290              **************************/
 291
 292             if (gmx_mm_any_lt(rsq20,rcutoff2))
 293             {
 294
 295             /* Compute parameters for interactions between i and j atoms */
 296             qq20             = _mm_mul_pd(iq2,jq0);
 297
 298             /* REACTION-FIELD ELECTROSTATICS */
 299             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 300             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 301
 302             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 303
 304             /* Update potential sum for this i atom from the interaction with this j atom. */
 305             velec            = _mm_and_pd(velec,cutoff_mask);
 306             velecsum         = _mm_add_pd(velecsum,velec);
 307
 308             fscal            = felec;
 309
 310             fscal            = _mm_and_pd(fscal,cutoff_mask);
 311
 312             /* Calculate temporary vectorial force */
 313             tx               = _mm_mul_pd(fscal,dx20);
 314             ty               = _mm_mul_pd(fscal,dy20);
 315             tz               = _mm_mul_pd(fscal,dz20);
 316
 317             /* Update vectorial force */
 318             fix2             = _mm_add_pd(fix2,tx);
 319             fiy2             = _mm_add_pd(fiy2,ty);
 320             fiz2             = _mm_add_pd(fiz2,tz);
 321
 322             fjx0             = _mm_add_pd(fjx0,tx);
 323             fjy0             = _mm_add_pd(fjy0,ty);
 324             fjz0             = _mm_add_pd(fjz0,tz);
 325
 326             }
 327
 328             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 329
 330             /* Inner loop uses 111 flops */
 331         }
 332
 333         if(jidx<j_index_end)
 334         {
 335
 336             jnrA             = jjnr[jidx];
 337             j_coord_offsetA  = DIM*jnrA;
 338
 339             /* load j atom coordinates */
 340             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 341                                               &jx0,&jy0,&jz0);
 342
 343             /* Calculate displacement vector */
 344             dx00             = _mm_sub_pd(ix0,jx0);
 345             dy00             = _mm_sub_pd(iy0,jy0);
 346             dz00             = _mm_sub_pd(iz0,jz0);
 347             dx10             = _mm_sub_pd(ix1,jx0);
 348             dy10             = _mm_sub_pd(iy1,jy0);
 349             dz10             = _mm_sub_pd(iz1,jz0);
 350             dx20             = _mm_sub_pd(ix2,jx0);
 351             dy20             = _mm_sub_pd(iy2,jy0);
 352             dz20             = _mm_sub_pd(iz2,jz0);
 353
 354             /* Calculate squared distance and things based on it */
 355             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 356             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 357             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 358
 359             rinv00           = sse41_invsqrt_d(rsq00);
 360             rinv10           = sse41_invsqrt_d(rsq10);
 361             rinv20           = sse41_invsqrt_d(rsq20);
 362
 363             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 364             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 365             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 366
 367             /* Load parameters for j particles */
 368             jq0              = _mm_load_sd(charge+jnrA+0);
 369
 370             fjx0             = _mm_setzero_pd();
 371             fjy0             = _mm_setzero_pd();
 372             fjz0             = _mm_setzero_pd();
 373
 374             /**************************
 375              * CALCULATE INTERACTIONS *
 376              **************************/
 377
 378             if (gmx_mm_any_lt(rsq00,rcutoff2))
 379             {
 380
 381             /* Compute parameters for interactions between i and j atoms */
 382             qq00             = _mm_mul_pd(iq0,jq0);
 383
 384             /* REACTION-FIELD ELECTROSTATICS */
 385             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 386             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 387
 388             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 389
 390             /* Update potential sum for this i atom from the interaction with this j atom. */
 391             velec            = _mm_and_pd(velec,cutoff_mask);
 392             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 393             velecsum         = _mm_add_pd(velecsum,velec);
 394
 395             fscal            = felec;
 396
 397             fscal            = _mm_and_pd(fscal,cutoff_mask);
 398
 399             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 400
 401             /* Calculate temporary vectorial force */
 402             tx               = _mm_mul_pd(fscal,dx00);
 403             ty               = _mm_mul_pd(fscal,dy00);
 404             tz               = _mm_mul_pd(fscal,dz00);
 405
 406             /* Update vectorial force */
 407             fix0             = _mm_add_pd(fix0,tx);
 408             fiy0             = _mm_add_pd(fiy0,ty);
 409             fiz0             = _mm_add_pd(fiz0,tz);
 410
 411             fjx0             = _mm_add_pd(fjx0,tx);
 412             fjy0             = _mm_add_pd(fjy0,ty);
 413             fjz0             = _mm_add_pd(fjz0,tz);
 414
 415             }
 416
 417             /**************************
 418              * CALCULATE INTERACTIONS *
 419              **************************/
 420
 421             if (gmx_mm_any_lt(rsq10,rcutoff2))
 422             {
 423
 424             /* Compute parameters for interactions between i and j atoms */
 425             qq10             = _mm_mul_pd(iq1,jq0);
 426
 427             /* REACTION-FIELD ELECTROSTATICS */
 428             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 429             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 430
 431             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 432
 433             /* Update potential sum for this i atom from the interaction with this j atom. */
 434             velec            = _mm_and_pd(velec,cutoff_mask);
 435             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 436             velecsum         = _mm_add_pd(velecsum,velec);
 437
 438             fscal            = felec;
 439
 440             fscal            = _mm_and_pd(fscal,cutoff_mask);
 441
 442             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 443
 444             /* Calculate temporary vectorial force */
 445             tx               = _mm_mul_pd(fscal,dx10);
 446             ty               = _mm_mul_pd(fscal,dy10);
 447             tz               = _mm_mul_pd(fscal,dz10);
 448
 449             /* Update vectorial force */
 450             fix1             = _mm_add_pd(fix1,tx);
 451             fiy1             = _mm_add_pd(fiy1,ty);
 452             fiz1             = _mm_add_pd(fiz1,tz);
 453
 454             fjx0             = _mm_add_pd(fjx0,tx);
 455             fjy0             = _mm_add_pd(fjy0,ty);
 456             fjz0             = _mm_add_pd(fjz0,tz);
 457
 458             }
 459
 460             /**************************
 461              * CALCULATE INTERACTIONS *
 462              **************************/
 463
 464             if (gmx_mm_any_lt(rsq20,rcutoff2))
 465             {
 466
 467             /* Compute parameters for interactions between i and j atoms */
 468             qq20             = _mm_mul_pd(iq2,jq0);
 469
 470             /* REACTION-FIELD ELECTROSTATICS */
 471             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 472             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 473
 474             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 475
 476             /* Update potential sum for this i atom from the interaction with this j atom. */
 477             velec            = _mm_and_pd(velec,cutoff_mask);
 478             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 479             velecsum         = _mm_add_pd(velecsum,velec);
 480
 481             fscal            = felec;
 482
 483             fscal            = _mm_and_pd(fscal,cutoff_mask);
 484
 485             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 486
 487             /* Calculate temporary vectorial force */
 488             tx               = _mm_mul_pd(fscal,dx20);
 489             ty               = _mm_mul_pd(fscal,dy20);
 490             tz               = _mm_mul_pd(fscal,dz20);
 491
 492             /* Update vectorial force */
 493             fix2             = _mm_add_pd(fix2,tx);
 494             fiy2             = _mm_add_pd(fiy2,ty);
 495             fiz2             = _mm_add_pd(fiz2,tz);
 496
 497             fjx0             = _mm_add_pd(fjx0,tx);
 498             fjy0             = _mm_add_pd(fjy0,ty);
 499             fjz0             = _mm_add_pd(fjz0,tz);
 500
 501             }
 502
 503             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 504
 505             /* Inner loop uses 111 flops */
 506         }
 507
 508         /* End of innermost loop */
 509
 510         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 511                                               f+i_coord_offset,fshift+i_shift_offset);
 512
 513         ggid                        = gid[iidx];
 514         /* Update potential energies */
 515         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 516
 517         /* Increment number of inner iterations */
 518         inneriter                  += j_index_end - j_index_start;
 519
 520         /* Outer loop uses 19 flops */
 521     }
 522
 523     /* Increment number of outer iterations */
 524     outeriter        += nri;
 525
 526     /* Update outer/inner flops */
 527
 528     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*111);
 529 }
 530 /*
 531  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomW3P1_F_sse4_1_double
 532  * Electrostatics interaction: ReactionField
 533  * VdW interaction:            None
 534  * Geometry:                   Water3-Particle
 535  * Calculate force/pot:        Force
 536  */
 537 void
 538 nb_kernel_ElecRFCut_VdwNone_GeomW3P1_F_sse4_1_double
 539                     (t_nblist                    * gmx_restrict       nlist,
 540                      rvec                        * gmx_restrict          xx,
 541                      rvec                        * gmx_restrict          ff,
 542                      struct t_forcerec           * gmx_restrict          fr,
 543                      t_mdatoms                   * gmx_restrict     mdatoms,
 544                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 545                      t_nrnb                      * gmx_restrict        nrnb)
 546 {
 547     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 548      * just 0 for non-waters.
 549      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 550      * jnr indices corresponding to data put in the four positions in the SIMD register.
 551      */
 552     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 553     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 554     int              jnrA,jnrB;
 555     int              j_coord_offsetA,j_coord_offsetB;
 556     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 557     real             rcutoff_scalar;
 558     real             *shiftvec,*fshift,*x,*f;
 559     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 560     int              vdwioffset0;
 561     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 562     int              vdwioffset1;
 563     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 564     int              vdwioffset2;
 565     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 566     int              vdwjidx0A,vdwjidx0B;
 567     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 568     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 569     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 570     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 571     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 572     real             *charge;
 573     __m128d          dummy_mask,cutoff_mask;
 574     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 575     __m128d          one     = _mm_set1_pd(1.0);
 576     __m128d          two     = _mm_set1_pd(2.0);
 577     x                = xx[0];
 578     f                = ff[0];
 579
 580     nri              = nlist->nri;
 581     iinr             = nlist->iinr;
 582     jindex           = nlist->jindex;
 583     jjnr             = nlist->jjnr;
 584     shiftidx         = nlist->shift;
 585     gid              = nlist->gid;
 586     shiftvec         = fr->shift_vec[0];
 587     fshift           = fr->fshift[0];
 588     facel            = _mm_set1_pd(fr->ic->epsfac);
 589     charge           = mdatoms->chargeA;
 590     krf              = _mm_set1_pd(fr->ic->k_rf);
 591     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 592     crf              = _mm_set1_pd(fr->ic->c_rf);
 593
 594     /* Setup water-specific parameters */
 595     inr              = nlist->iinr[0];
 596     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 597     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 598     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 599
 600     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 601     rcutoff_scalar   = fr->ic->rcoulomb;
 602     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 603     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 604
 605     /* Avoid stupid compiler warnings */
 606     jnrA = jnrB = 0;
 607     j_coord_offsetA = 0;
 608     j_coord_offsetB = 0;
 609
 610     outeriter        = 0;
 611     inneriter        = 0;
 612
 613     /* Start outer loop over neighborlists */
 614     for(iidx=0; iidx<nri; iidx++)
 615     {
 616         /* Load shift vector for this list */
 617         i_shift_offset   = DIM*shiftidx[iidx];
 618
 619         /* Load limits for loop over neighbors */
 620         j_index_start    = jindex[iidx];
 621         j_index_end      = jindex[iidx+1];
 622
 623         /* Get outer coordinate index */
 624         inr              = iinr[iidx];
 625         i_coord_offset   = DIM*inr;
 626
 627         /* Load i particle coords and add shift vector */
 628         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 629                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 630
 631         fix0             = _mm_setzero_pd();
 632         fiy0             = _mm_setzero_pd();
 633         fiz0             = _mm_setzero_pd();
 634         fix1             = _mm_setzero_pd();
 635         fiy1             = _mm_setzero_pd();
 636         fiz1             = _mm_setzero_pd();
 637         fix2             = _mm_setzero_pd();
 638         fiy2             = _mm_setzero_pd();
 639         fiz2             = _mm_setzero_pd();
 640
 641         /* Start inner kernel loop */
 642         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 643         {
 644
 645             /* Get j neighbor index, and coordinate index */
 646             jnrA             = jjnr[jidx];
 647             jnrB             = jjnr[jidx+1];
 648             j_coord_offsetA  = DIM*jnrA;
 649             j_coord_offsetB  = DIM*jnrB;
 650
 651             /* load j atom coordinates */
 652             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 653                                               &jx0,&jy0,&jz0);
 654
 655             /* Calculate displacement vector */
 656             dx00             = _mm_sub_pd(ix0,jx0);
 657             dy00             = _mm_sub_pd(iy0,jy0);
 658             dz00             = _mm_sub_pd(iz0,jz0);
 659             dx10             = _mm_sub_pd(ix1,jx0);
 660             dy10             = _mm_sub_pd(iy1,jy0);
 661             dz10             = _mm_sub_pd(iz1,jz0);
 662             dx20             = _mm_sub_pd(ix2,jx0);
 663             dy20             = _mm_sub_pd(iy2,jy0);
 664             dz20             = _mm_sub_pd(iz2,jz0);
 665
 666             /* Calculate squared distance and things based on it */
 667             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 668             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 669             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 670
 671             rinv00           = sse41_invsqrt_d(rsq00);
 672             rinv10           = sse41_invsqrt_d(rsq10);
 673             rinv20           = sse41_invsqrt_d(rsq20);
 674
 675             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 676             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 677             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 678
 679             /* Load parameters for j particles */
 680             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 681
 682             fjx0             = _mm_setzero_pd();
 683             fjy0             = _mm_setzero_pd();
 684             fjz0             = _mm_setzero_pd();
 685
 686             /**************************
 687              * CALCULATE INTERACTIONS *
 688              **************************/
 689
 690             if (gmx_mm_any_lt(rsq00,rcutoff2))
 691             {
 692
 693             /* Compute parameters for interactions between i and j atoms */
 694             qq00             = _mm_mul_pd(iq0,jq0);
 695
 696             /* REACTION-FIELD ELECTROSTATICS */
 697             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 698
 699             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 700
 701             fscal            = felec;
 702
 703             fscal            = _mm_and_pd(fscal,cutoff_mask);
 704
 705             /* Calculate temporary vectorial force */
 706             tx               = _mm_mul_pd(fscal,dx00);
 707             ty               = _mm_mul_pd(fscal,dy00);
 708             tz               = _mm_mul_pd(fscal,dz00);
 709
 710             /* Update vectorial force */
 711             fix0             = _mm_add_pd(fix0,tx);
 712             fiy0             = _mm_add_pd(fiy0,ty);
 713             fiz0             = _mm_add_pd(fiz0,tz);
 714
 715             fjx0             = _mm_add_pd(fjx0,tx);
 716             fjy0             = _mm_add_pd(fjy0,ty);
 717             fjz0             = _mm_add_pd(fjz0,tz);
 718
 719             }
 720
 721             /**************************
 722              * CALCULATE INTERACTIONS *
 723              **************************/
 724
 725             if (gmx_mm_any_lt(rsq10,rcutoff2))
 726             {
 727
 728             /* Compute parameters for interactions between i and j atoms */
 729             qq10             = _mm_mul_pd(iq1,jq0);
 730
 731             /* REACTION-FIELD ELECTROSTATICS */
 732             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 733
 734             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 735
 736             fscal            = felec;
 737
 738             fscal            = _mm_and_pd(fscal,cutoff_mask);
 739
 740             /* Calculate temporary vectorial force */
 741             tx               = _mm_mul_pd(fscal,dx10);
 742             ty               = _mm_mul_pd(fscal,dy10);
 743             tz               = _mm_mul_pd(fscal,dz10);
 744
 745             /* Update vectorial force */
 746             fix1             = _mm_add_pd(fix1,tx);
 747             fiy1             = _mm_add_pd(fiy1,ty);
 748             fiz1             = _mm_add_pd(fiz1,tz);
 749
 750             fjx0             = _mm_add_pd(fjx0,tx);
 751             fjy0             = _mm_add_pd(fjy0,ty);
 752             fjz0             = _mm_add_pd(fjz0,tz);
 753
 754             }
 755
 756             /**************************
 757              * CALCULATE INTERACTIONS *
 758              **************************/
 759
 760             if (gmx_mm_any_lt(rsq20,rcutoff2))
 761             {
 762
 763             /* Compute parameters for interactions between i and j atoms */
 764             qq20             = _mm_mul_pd(iq2,jq0);
 765
 766             /* REACTION-FIELD ELECTROSTATICS */
 767             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 768
 769             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 770
 771             fscal            = felec;
 772
 773             fscal            = _mm_and_pd(fscal,cutoff_mask);
 774
 775             /* Calculate temporary vectorial force */
 776             tx               = _mm_mul_pd(fscal,dx20);
 777             ty               = _mm_mul_pd(fscal,dy20);
 778             tz               = _mm_mul_pd(fscal,dz20);
 779
 780             /* Update vectorial force */
 781             fix2             = _mm_add_pd(fix2,tx);
 782             fiy2             = _mm_add_pd(fiy2,ty);
 783             fiz2             = _mm_add_pd(fiz2,tz);
 784
 785             fjx0             = _mm_add_pd(fjx0,tx);
 786             fjy0             = _mm_add_pd(fjy0,ty);
 787             fjz0             = _mm_add_pd(fjz0,tz);
 788
 789             }
 790
 791             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 792
 793             /* Inner loop uses 93 flops */
 794         }
 795
 796         if(jidx<j_index_end)
 797         {
 798
 799             jnrA             = jjnr[jidx];
 800             j_coord_offsetA  = DIM*jnrA;
 801
 802             /* load j atom coordinates */
 803             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 804                                               &jx0,&jy0,&jz0);
 805
 806             /* Calculate displacement vector */
 807             dx00             = _mm_sub_pd(ix0,jx0);
 808             dy00             = _mm_sub_pd(iy0,jy0);
 809             dz00             = _mm_sub_pd(iz0,jz0);
 810             dx10             = _mm_sub_pd(ix1,jx0);
 811             dy10             = _mm_sub_pd(iy1,jy0);
 812             dz10             = _mm_sub_pd(iz1,jz0);
 813             dx20             = _mm_sub_pd(ix2,jx0);
 814             dy20             = _mm_sub_pd(iy2,jy0);
 815             dz20             = _mm_sub_pd(iz2,jz0);
 816
 817             /* Calculate squared distance and things based on it */
 818             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 819             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 820             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 821
 822             rinv00           = sse41_invsqrt_d(rsq00);
 823             rinv10           = sse41_invsqrt_d(rsq10);
 824             rinv20           = sse41_invsqrt_d(rsq20);
 825
 826             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 827             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 828             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 829
 830             /* Load parameters for j particles */
 831             jq0              = _mm_load_sd(charge+jnrA+0);
 832
 833             fjx0             = _mm_setzero_pd();
 834             fjy0             = _mm_setzero_pd();
 835             fjz0             = _mm_setzero_pd();
 836
 837             /**************************
 838              * CALCULATE INTERACTIONS *
 839              **************************/
 840
 841             if (gmx_mm_any_lt(rsq00,rcutoff2))
 842             {
 843
 844             /* Compute parameters for interactions between i and j atoms */
 845             qq00             = _mm_mul_pd(iq0,jq0);
 846
 847             /* REACTION-FIELD ELECTROSTATICS */
 848             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 849
 850             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 851
 852             fscal            = felec;
 853
 854             fscal            = _mm_and_pd(fscal,cutoff_mask);
 855
 856             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 857
 858             /* Calculate temporary vectorial force */
 859             tx               = _mm_mul_pd(fscal,dx00);
 860             ty               = _mm_mul_pd(fscal,dy00);
 861             tz               = _mm_mul_pd(fscal,dz00);
 862
 863             /* Update vectorial force */
 864             fix0             = _mm_add_pd(fix0,tx);
 865             fiy0             = _mm_add_pd(fiy0,ty);
 866             fiz0             = _mm_add_pd(fiz0,tz);
 867
 868             fjx0             = _mm_add_pd(fjx0,tx);
 869             fjy0             = _mm_add_pd(fjy0,ty);
 870             fjz0             = _mm_add_pd(fjz0,tz);
 871
 872             }
 873
 874             /**************************
 875              * CALCULATE INTERACTIONS *
 876              **************************/
 877
 878             if (gmx_mm_any_lt(rsq10,rcutoff2))
 879             {
 880
 881             /* Compute parameters for interactions between i and j atoms */
 882             qq10             = _mm_mul_pd(iq1,jq0);
 883
 884             /* REACTION-FIELD ELECTROSTATICS */
 885             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 886
 887             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 888
 889             fscal            = felec;
 890
 891             fscal            = _mm_and_pd(fscal,cutoff_mask);
 892
 893             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 894
 895             /* Calculate temporary vectorial force */
 896             tx               = _mm_mul_pd(fscal,dx10);
 897             ty               = _mm_mul_pd(fscal,dy10);
 898             tz               = _mm_mul_pd(fscal,dz10);
 899
 900             /* Update vectorial force */
 901             fix1             = _mm_add_pd(fix1,tx);
 902             fiy1             = _mm_add_pd(fiy1,ty);
 903             fiz1             = _mm_add_pd(fiz1,tz);
 904
 905             fjx0             = _mm_add_pd(fjx0,tx);
 906             fjy0             = _mm_add_pd(fjy0,ty);
 907             fjz0             = _mm_add_pd(fjz0,tz);
 908
 909             }
 910
 911             /**************************
 912              * CALCULATE INTERACTIONS *
 913              **************************/
 914
 915             if (gmx_mm_any_lt(rsq20,rcutoff2))
 916             {
 917
 918             /* Compute parameters for interactions between i and j atoms */
 919             qq20             = _mm_mul_pd(iq2,jq0);
 920
 921             /* REACTION-FIELD ELECTROSTATICS */
 922             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 923
 924             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 925
 926             fscal            = felec;
 927
 928             fscal            = _mm_and_pd(fscal,cutoff_mask);
 929
 930             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 931
 932             /* Calculate temporary vectorial force */
 933             tx               = _mm_mul_pd(fscal,dx20);
 934             ty               = _mm_mul_pd(fscal,dy20);
 935             tz               = _mm_mul_pd(fscal,dz20);
 936
 937             /* Update vectorial force */
 938             fix2             = _mm_add_pd(fix2,tx);
 939             fiy2             = _mm_add_pd(fiy2,ty);
 940             fiz2             = _mm_add_pd(fiz2,tz);
 941
 942             fjx0             = _mm_add_pd(fjx0,tx);
 943             fjy0             = _mm_add_pd(fjy0,ty);
 944             fjz0             = _mm_add_pd(fjz0,tz);
 945
 946             }
 947
 948             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 949
 950             /* Inner loop uses 93 flops */
 951         }
 952
 953         /* End of innermost loop */
 954
 955         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 956                                               f+i_coord_offset,fshift+i_shift_offset);
 957
 958         /* Increment number of inner iterations */
 959         inneriter                  += j_index_end - j_index_start;
 960
 961         /* Outer loop uses 18 flops */
 962     }
 963
 964     /* Increment number of outer iterations */
 965     outeriter        += nri;
 966
 967     /* Update outer/inner flops */
 968
 969     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*93);
 970 }