src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecCoul_VdwNone_GeomW3P1_sse2_single.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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 sse2_single 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_sse2_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomW3P1_VF_sse2_single
  51  * Electrostatics interaction: Coulomb
  52  * VdW interaction:            None
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCoul_VdwNone_GeomW3P1_VF_sse2_single
  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,C,D refer to j loop unrolling done with SSE, e.g. for the four 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,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  89     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     __m128           dummy_mask,cutoff_mask;
  96     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  97     __m128           one     = _mm_set1_ps(1.0);
  98     __m128           two     = _mm_set1_ps(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_ps(fr->ic->epsfac);
 111     charge           = mdatoms->chargeA;
 112
 113     /* Setup water-specific parameters */
 114     inr              = nlist->iinr[0];
 115     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 116     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 117     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 118
 119     /* Avoid stupid compiler warnings */
 120     jnrA = jnrB = jnrC = jnrD = 0;
 121     j_coord_offsetA = 0;
 122     j_coord_offsetB = 0;
 123     j_coord_offsetC = 0;
 124     j_coord_offsetD = 0;
 125
 126     outeriter        = 0;
 127     inneriter        = 0;
 128
 129     for(iidx=0;iidx<4*DIM;iidx++)
 130     {
 131         scratch[iidx] = 0.0;
 132     }
 133
 134     /* Start outer loop over neighborlists */
 135     for(iidx=0; iidx<nri; iidx++)
 136     {
 137         /* Load shift vector for this list */
 138         i_shift_offset   = DIM*shiftidx[iidx];
 139
 140         /* Load limits for loop over neighbors */
 141         j_index_start    = jindex[iidx];
 142         j_index_end      = jindex[iidx+1];
 143
 144         /* Get outer coordinate index */
 145         inr              = iinr[iidx];
 146         i_coord_offset   = DIM*inr;
 147
 148         /* Load i particle coords and add shift vector */
 149         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 150                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 151
 152         fix0             = _mm_setzero_ps();
 153         fiy0             = _mm_setzero_ps();
 154         fiz0             = _mm_setzero_ps();
 155         fix1             = _mm_setzero_ps();
 156         fiy1             = _mm_setzero_ps();
 157         fiz1             = _mm_setzero_ps();
 158         fix2             = _mm_setzero_ps();
 159         fiy2             = _mm_setzero_ps();
 160         fiz2             = _mm_setzero_ps();
 161
 162         /* Reset potential sums */
 163         velecsum         = _mm_setzero_ps();
 164
 165         /* Start inner kernel loop */
 166         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 167         {
 168
 169             /* Get j neighbor index, and coordinate index */
 170             jnrA             = jjnr[jidx];
 171             jnrB             = jjnr[jidx+1];
 172             jnrC             = jjnr[jidx+2];
 173             jnrD             = jjnr[jidx+3];
 174             j_coord_offsetA  = DIM*jnrA;
 175             j_coord_offsetB  = DIM*jnrB;
 176             j_coord_offsetC  = DIM*jnrC;
 177             j_coord_offsetD  = DIM*jnrD;
 178
 179             /* load j atom coordinates */
 180             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 181                                               x+j_coord_offsetC,x+j_coord_offsetD,
 182                                               &jx0,&jy0,&jz0);
 183
 184             /* Calculate displacement vector */
 185             dx00             = _mm_sub_ps(ix0,jx0);
 186             dy00             = _mm_sub_ps(iy0,jy0);
 187             dz00             = _mm_sub_ps(iz0,jz0);
 188             dx10             = _mm_sub_ps(ix1,jx0);
 189             dy10             = _mm_sub_ps(iy1,jy0);
 190             dz10             = _mm_sub_ps(iz1,jz0);
 191             dx20             = _mm_sub_ps(ix2,jx0);
 192             dy20             = _mm_sub_ps(iy2,jy0);
 193             dz20             = _mm_sub_ps(iz2,jz0);
 194
 195             /* Calculate squared distance and things based on it */
 196             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 197             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 198             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 199
 200             rinv00           = sse2_invsqrt_f(rsq00);
 201             rinv10           = sse2_invsqrt_f(rsq10);
 202             rinv20           = sse2_invsqrt_f(rsq20);
 203
 204             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 205             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 206             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 207
 208             /* Load parameters for j particles */
 209             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 210                                                               charge+jnrC+0,charge+jnrD+0);
 211
 212             fjx0             = _mm_setzero_ps();
 213             fjy0             = _mm_setzero_ps();
 214             fjz0             = _mm_setzero_ps();
 215
 216             /**************************
 217              * CALCULATE INTERACTIONS *
 218              **************************/
 219
 220             /* Compute parameters for interactions between i and j atoms */
 221             qq00             = _mm_mul_ps(iq0,jq0);
 222
 223             /* COULOMB ELECTROSTATICS */
 224             velec            = _mm_mul_ps(qq00,rinv00);
 225             felec            = _mm_mul_ps(velec,rinvsq00);
 226
 227             /* Update potential sum for this i atom from the interaction with this j atom. */
 228             velecsum         = _mm_add_ps(velecsum,velec);
 229
 230             fscal            = felec;
 231
 232             /* Calculate temporary vectorial force */
 233             tx               = _mm_mul_ps(fscal,dx00);
 234             ty               = _mm_mul_ps(fscal,dy00);
 235             tz               = _mm_mul_ps(fscal,dz00);
 236
 237             /* Update vectorial force */
 238             fix0             = _mm_add_ps(fix0,tx);
 239             fiy0             = _mm_add_ps(fiy0,ty);
 240             fiz0             = _mm_add_ps(fiz0,tz);
 241
 242             fjx0             = _mm_add_ps(fjx0,tx);
 243             fjy0             = _mm_add_ps(fjy0,ty);
 244             fjz0             = _mm_add_ps(fjz0,tz);
 245
 246             /**************************
 247              * CALCULATE INTERACTIONS *
 248              **************************/
 249
 250             /* Compute parameters for interactions between i and j atoms */
 251             qq10             = _mm_mul_ps(iq1,jq0);
 252
 253             /* COULOMB ELECTROSTATICS */
 254             velec            = _mm_mul_ps(qq10,rinv10);
 255             felec            = _mm_mul_ps(velec,rinvsq10);
 256
 257             /* Update potential sum for this i atom from the interaction with this j atom. */
 258             velecsum         = _mm_add_ps(velecsum,velec);
 259
 260             fscal            = felec;
 261
 262             /* Calculate temporary vectorial force */
 263             tx               = _mm_mul_ps(fscal,dx10);
 264             ty               = _mm_mul_ps(fscal,dy10);
 265             tz               = _mm_mul_ps(fscal,dz10);
 266
 267             /* Update vectorial force */
 268             fix1             = _mm_add_ps(fix1,tx);
 269             fiy1             = _mm_add_ps(fiy1,ty);
 270             fiz1             = _mm_add_ps(fiz1,tz);
 271
 272             fjx0             = _mm_add_ps(fjx0,tx);
 273             fjy0             = _mm_add_ps(fjy0,ty);
 274             fjz0             = _mm_add_ps(fjz0,tz);
 275
 276             /**************************
 277              * CALCULATE INTERACTIONS *
 278              **************************/
 279
 280             /* Compute parameters for interactions between i and j atoms */
 281             qq20             = _mm_mul_ps(iq2,jq0);
 282
 283             /* COULOMB ELECTROSTATICS */
 284             velec            = _mm_mul_ps(qq20,rinv20);
 285             felec            = _mm_mul_ps(velec,rinvsq20);
 286
 287             /* Update potential sum for this i atom from the interaction with this j atom. */
 288             velecsum         = _mm_add_ps(velecsum,velec);
 289
 290             fscal            = felec;
 291
 292             /* Calculate temporary vectorial force */
 293             tx               = _mm_mul_ps(fscal,dx20);
 294             ty               = _mm_mul_ps(fscal,dy20);
 295             tz               = _mm_mul_ps(fscal,dz20);
 296
 297             /* Update vectorial force */
 298             fix2             = _mm_add_ps(fix2,tx);
 299             fiy2             = _mm_add_ps(fiy2,ty);
 300             fiz2             = _mm_add_ps(fiz2,tz);
 301
 302             fjx0             = _mm_add_ps(fjx0,tx);
 303             fjy0             = _mm_add_ps(fjy0,ty);
 304             fjz0             = _mm_add_ps(fjz0,tz);
 305
 306             fjptrA             = f+j_coord_offsetA;
 307             fjptrB             = f+j_coord_offsetB;
 308             fjptrC             = f+j_coord_offsetC;
 309             fjptrD             = f+j_coord_offsetD;
 310
 311             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 312
 313             /* Inner loop uses 84 flops */
 314         }
 315
 316         if(jidx<j_index_end)
 317         {
 318
 319             /* Get j neighbor index, and coordinate index */
 320             jnrlistA         = jjnr[jidx];
 321             jnrlistB         = jjnr[jidx+1];
 322             jnrlistC         = jjnr[jidx+2];
 323             jnrlistD         = jjnr[jidx+3];
 324             /* Sign of each element will be negative for non-real atoms.
 325              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 326              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 327              */
 328             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 329             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 330             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 331             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 332             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 333             j_coord_offsetA  = DIM*jnrA;
 334             j_coord_offsetB  = DIM*jnrB;
 335             j_coord_offsetC  = DIM*jnrC;
 336             j_coord_offsetD  = DIM*jnrD;
 337
 338             /* load j atom coordinates */
 339             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 340                                               x+j_coord_offsetC,x+j_coord_offsetD,
 341                                               &jx0,&jy0,&jz0);
 342
 343             /* Calculate displacement vector */
 344             dx00             = _mm_sub_ps(ix0,jx0);
 345             dy00             = _mm_sub_ps(iy0,jy0);
 346             dz00             = _mm_sub_ps(iz0,jz0);
 347             dx10             = _mm_sub_ps(ix1,jx0);
 348             dy10             = _mm_sub_ps(iy1,jy0);
 349             dz10             = _mm_sub_ps(iz1,jz0);
 350             dx20             = _mm_sub_ps(ix2,jx0);
 351             dy20             = _mm_sub_ps(iy2,jy0);
 352             dz20             = _mm_sub_ps(iz2,jz0);
 353
 354             /* Calculate squared distance and things based on it */
 355             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 356             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 357             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 358
 359             rinv00           = sse2_invsqrt_f(rsq00);
 360             rinv10           = sse2_invsqrt_f(rsq10);
 361             rinv20           = sse2_invsqrt_f(rsq20);
 362
 363             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 364             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 365             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 366
 367             /* Load parameters for j particles */
 368             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 369                                                               charge+jnrC+0,charge+jnrD+0);
 370
 371             fjx0             = _mm_setzero_ps();
 372             fjy0             = _mm_setzero_ps();
 373             fjz0             = _mm_setzero_ps();
 374
 375             /**************************
 376              * CALCULATE INTERACTIONS *
 377              **************************/
 378
 379             /* Compute parameters for interactions between i and j atoms */
 380             qq00             = _mm_mul_ps(iq0,jq0);
 381
 382             /* COULOMB ELECTROSTATICS */
 383             velec            = _mm_mul_ps(qq00,rinv00);
 384             felec            = _mm_mul_ps(velec,rinvsq00);
 385
 386             /* Update potential sum for this i atom from the interaction with this j atom. */
 387             velec            = _mm_andnot_ps(dummy_mask,velec);
 388             velecsum         = _mm_add_ps(velecsum,velec);
 389
 390             fscal            = felec;
 391
 392             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 393
 394             /* Calculate temporary vectorial force */
 395             tx               = _mm_mul_ps(fscal,dx00);
 396             ty               = _mm_mul_ps(fscal,dy00);
 397             tz               = _mm_mul_ps(fscal,dz00);
 398
 399             /* Update vectorial force */
 400             fix0             = _mm_add_ps(fix0,tx);
 401             fiy0             = _mm_add_ps(fiy0,ty);
 402             fiz0             = _mm_add_ps(fiz0,tz);
 403
 404             fjx0             = _mm_add_ps(fjx0,tx);
 405             fjy0             = _mm_add_ps(fjy0,ty);
 406             fjz0             = _mm_add_ps(fjz0,tz);
 407
 408             /**************************
 409              * CALCULATE INTERACTIONS *
 410              **************************/
 411
 412             /* Compute parameters for interactions between i and j atoms */
 413             qq10             = _mm_mul_ps(iq1,jq0);
 414
 415             /* COULOMB ELECTROSTATICS */
 416             velec            = _mm_mul_ps(qq10,rinv10);
 417             felec            = _mm_mul_ps(velec,rinvsq10);
 418
 419             /* Update potential sum for this i atom from the interaction with this j atom. */
 420             velec            = _mm_andnot_ps(dummy_mask,velec);
 421             velecsum         = _mm_add_ps(velecsum,velec);
 422
 423             fscal            = felec;
 424
 425             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 426
 427             /* Calculate temporary vectorial force */
 428             tx               = _mm_mul_ps(fscal,dx10);
 429             ty               = _mm_mul_ps(fscal,dy10);
 430             tz               = _mm_mul_ps(fscal,dz10);
 431
 432             /* Update vectorial force */
 433             fix1             = _mm_add_ps(fix1,tx);
 434             fiy1             = _mm_add_ps(fiy1,ty);
 435             fiz1             = _mm_add_ps(fiz1,tz);
 436
 437             fjx0             = _mm_add_ps(fjx0,tx);
 438             fjy0             = _mm_add_ps(fjy0,ty);
 439             fjz0             = _mm_add_ps(fjz0,tz);
 440
 441             /**************************
 442              * CALCULATE INTERACTIONS *
 443              **************************/
 444
 445             /* Compute parameters for interactions between i and j atoms */
 446             qq20             = _mm_mul_ps(iq2,jq0);
 447
 448             /* COULOMB ELECTROSTATICS */
 449             velec            = _mm_mul_ps(qq20,rinv20);
 450             felec            = _mm_mul_ps(velec,rinvsq20);
 451
 452             /* Update potential sum for this i atom from the interaction with this j atom. */
 453             velec            = _mm_andnot_ps(dummy_mask,velec);
 454             velecsum         = _mm_add_ps(velecsum,velec);
 455
 456             fscal            = felec;
 457
 458             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 459
 460             /* Calculate temporary vectorial force */
 461             tx               = _mm_mul_ps(fscal,dx20);
 462             ty               = _mm_mul_ps(fscal,dy20);
 463             tz               = _mm_mul_ps(fscal,dz20);
 464
 465             /* Update vectorial force */
 466             fix2             = _mm_add_ps(fix2,tx);
 467             fiy2             = _mm_add_ps(fiy2,ty);
 468             fiz2             = _mm_add_ps(fiz2,tz);
 469
 470             fjx0             = _mm_add_ps(fjx0,tx);
 471             fjy0             = _mm_add_ps(fjy0,ty);
 472             fjz0             = _mm_add_ps(fjz0,tz);
 473
 474             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 475             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 476             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 477             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 478
 479             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 480
 481             /* Inner loop uses 84 flops */
 482         }
 483
 484         /* End of innermost loop */
 485
 486         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 487                                               f+i_coord_offset,fshift+i_shift_offset);
 488
 489         ggid                        = gid[iidx];
 490         /* Update potential energies */
 491         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 492
 493         /* Increment number of inner iterations */
 494         inneriter                  += j_index_end - j_index_start;
 495
 496         /* Outer loop uses 19 flops */
 497     }
 498
 499     /* Increment number of outer iterations */
 500     outeriter        += nri;
 501
 502     /* Update outer/inner flops */
 503
 504     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*84);
 505 }
 506 /*
 507  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomW3P1_F_sse2_single
 508  * Electrostatics interaction: Coulomb
 509  * VdW interaction:            None
 510  * Geometry:                   Water3-Particle
 511  * Calculate force/pot:        Force
 512  */
 513 void
 514 nb_kernel_ElecCoul_VdwNone_GeomW3P1_F_sse2_single
 515                     (t_nblist                    * gmx_restrict       nlist,
 516                      rvec                        * gmx_restrict          xx,
 517                      rvec                        * gmx_restrict          ff,
 518                      struct t_forcerec           * gmx_restrict          fr,
 519                      t_mdatoms                   * gmx_restrict     mdatoms,
 520                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 521                      t_nrnb                      * gmx_restrict        nrnb)
 522 {
 523     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 524      * just 0 for non-waters.
 525      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 526      * jnr indices corresponding to data put in the four positions in the SIMD register.
 527      */
 528     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 529     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 530     int              jnrA,jnrB,jnrC,jnrD;
 531     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 532     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 533     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 534     real             rcutoff_scalar;
 535     real             *shiftvec,*fshift,*x,*f;
 536     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 537     real             scratch[4*DIM];
 538     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 539     int              vdwioffset0;
 540     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 541     int              vdwioffset1;
 542     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 543     int              vdwioffset2;
 544     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 545     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 546     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 547     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 548     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 549     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 550     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 551     real             *charge;
 552     __m128           dummy_mask,cutoff_mask;
 553     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 554     __m128           one     = _mm_set1_ps(1.0);
 555     __m128           two     = _mm_set1_ps(2.0);
 556     x                = xx[0];
 557     f                = ff[0];
 558
 559     nri              = nlist->nri;
 560     iinr             = nlist->iinr;
 561     jindex           = nlist->jindex;
 562     jjnr             = nlist->jjnr;
 563     shiftidx         = nlist->shift;
 564     gid              = nlist->gid;
 565     shiftvec         = fr->shift_vec[0];
 566     fshift           = fr->fshift[0];
 567     facel            = _mm_set1_ps(fr->ic->epsfac);
 568     charge           = mdatoms->chargeA;
 569
 570     /* Setup water-specific parameters */
 571     inr              = nlist->iinr[0];
 572     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 573     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 574     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 575
 576     /* Avoid stupid compiler warnings */
 577     jnrA = jnrB = jnrC = jnrD = 0;
 578     j_coord_offsetA = 0;
 579     j_coord_offsetB = 0;
 580     j_coord_offsetC = 0;
 581     j_coord_offsetD = 0;
 582
 583     outeriter        = 0;
 584     inneriter        = 0;
 585
 586     for(iidx=0;iidx<4*DIM;iidx++)
 587     {
 588         scratch[iidx] = 0.0;
 589     }
 590
 591     /* Start outer loop over neighborlists */
 592     for(iidx=0; iidx<nri; iidx++)
 593     {
 594         /* Load shift vector for this list */
 595         i_shift_offset   = DIM*shiftidx[iidx];
 596
 597         /* Load limits for loop over neighbors */
 598         j_index_start    = jindex[iidx];
 599         j_index_end      = jindex[iidx+1];
 600
 601         /* Get outer coordinate index */
 602         inr              = iinr[iidx];
 603         i_coord_offset   = DIM*inr;
 604
 605         /* Load i particle coords and add shift vector */
 606         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 607                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 608
 609         fix0             = _mm_setzero_ps();
 610         fiy0             = _mm_setzero_ps();
 611         fiz0             = _mm_setzero_ps();
 612         fix1             = _mm_setzero_ps();
 613         fiy1             = _mm_setzero_ps();
 614         fiz1             = _mm_setzero_ps();
 615         fix2             = _mm_setzero_ps();
 616         fiy2             = _mm_setzero_ps();
 617         fiz2             = _mm_setzero_ps();
 618
 619         /* Start inner kernel loop */
 620         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 621         {
 622
 623             /* Get j neighbor index, and coordinate index */
 624             jnrA             = jjnr[jidx];
 625             jnrB             = jjnr[jidx+1];
 626             jnrC             = jjnr[jidx+2];
 627             jnrD             = jjnr[jidx+3];
 628             j_coord_offsetA  = DIM*jnrA;
 629             j_coord_offsetB  = DIM*jnrB;
 630             j_coord_offsetC  = DIM*jnrC;
 631             j_coord_offsetD  = DIM*jnrD;
 632
 633             /* load j atom coordinates */
 634             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 635                                               x+j_coord_offsetC,x+j_coord_offsetD,
 636                                               &jx0,&jy0,&jz0);
 637
 638             /* Calculate displacement vector */
 639             dx00             = _mm_sub_ps(ix0,jx0);
 640             dy00             = _mm_sub_ps(iy0,jy0);
 641             dz00             = _mm_sub_ps(iz0,jz0);
 642             dx10             = _mm_sub_ps(ix1,jx0);
 643             dy10             = _mm_sub_ps(iy1,jy0);
 644             dz10             = _mm_sub_ps(iz1,jz0);
 645             dx20             = _mm_sub_ps(ix2,jx0);
 646             dy20             = _mm_sub_ps(iy2,jy0);
 647             dz20             = _mm_sub_ps(iz2,jz0);
 648
 649             /* Calculate squared distance and things based on it */
 650             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 651             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 652             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 653
 654             rinv00           = sse2_invsqrt_f(rsq00);
 655             rinv10           = sse2_invsqrt_f(rsq10);
 656             rinv20           = sse2_invsqrt_f(rsq20);
 657
 658             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 659             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 660             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 661
 662             /* Load parameters for j particles */
 663             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 664                                                               charge+jnrC+0,charge+jnrD+0);
 665
 666             fjx0             = _mm_setzero_ps();
 667             fjy0             = _mm_setzero_ps();
 668             fjz0             = _mm_setzero_ps();
 669
 670             /**************************
 671              * CALCULATE INTERACTIONS *
 672              **************************/
 673
 674             /* Compute parameters for interactions between i and j atoms */
 675             qq00             = _mm_mul_ps(iq0,jq0);
 676
 677             /* COULOMB ELECTROSTATICS */
 678             velec            = _mm_mul_ps(qq00,rinv00);
 679             felec            = _mm_mul_ps(velec,rinvsq00);
 680
 681             fscal            = felec;
 682
 683             /* Calculate temporary vectorial force */
 684             tx               = _mm_mul_ps(fscal,dx00);
 685             ty               = _mm_mul_ps(fscal,dy00);
 686             tz               = _mm_mul_ps(fscal,dz00);
 687
 688             /* Update vectorial force */
 689             fix0             = _mm_add_ps(fix0,tx);
 690             fiy0             = _mm_add_ps(fiy0,ty);
 691             fiz0             = _mm_add_ps(fiz0,tz);
 692
 693             fjx0             = _mm_add_ps(fjx0,tx);
 694             fjy0             = _mm_add_ps(fjy0,ty);
 695             fjz0             = _mm_add_ps(fjz0,tz);
 696
 697             /**************************
 698              * CALCULATE INTERACTIONS *
 699              **************************/
 700
 701             /* Compute parameters for interactions between i and j atoms */
 702             qq10             = _mm_mul_ps(iq1,jq0);
 703
 704             /* COULOMB ELECTROSTATICS */
 705             velec            = _mm_mul_ps(qq10,rinv10);
 706             felec            = _mm_mul_ps(velec,rinvsq10);
 707
 708             fscal            = felec;
 709
 710             /* Calculate temporary vectorial force */
 711             tx               = _mm_mul_ps(fscal,dx10);
 712             ty               = _mm_mul_ps(fscal,dy10);
 713             tz               = _mm_mul_ps(fscal,dz10);
 714
 715             /* Update vectorial force */
 716             fix1             = _mm_add_ps(fix1,tx);
 717             fiy1             = _mm_add_ps(fiy1,ty);
 718             fiz1             = _mm_add_ps(fiz1,tz);
 719
 720             fjx0             = _mm_add_ps(fjx0,tx);
 721             fjy0             = _mm_add_ps(fjy0,ty);
 722             fjz0             = _mm_add_ps(fjz0,tz);
 723
 724             /**************************
 725              * CALCULATE INTERACTIONS *
 726              **************************/
 727
 728             /* Compute parameters for interactions between i and j atoms */
 729             qq20             = _mm_mul_ps(iq2,jq0);
 730
 731             /* COULOMB ELECTROSTATICS */
 732             velec            = _mm_mul_ps(qq20,rinv20);
 733             felec            = _mm_mul_ps(velec,rinvsq20);
 734
 735             fscal            = felec;
 736
 737             /* Calculate temporary vectorial force */
 738             tx               = _mm_mul_ps(fscal,dx20);
 739             ty               = _mm_mul_ps(fscal,dy20);
 740             tz               = _mm_mul_ps(fscal,dz20);
 741
 742             /* Update vectorial force */
 743             fix2             = _mm_add_ps(fix2,tx);
 744             fiy2             = _mm_add_ps(fiy2,ty);
 745             fiz2             = _mm_add_ps(fiz2,tz);
 746
 747             fjx0             = _mm_add_ps(fjx0,tx);
 748             fjy0             = _mm_add_ps(fjy0,ty);
 749             fjz0             = _mm_add_ps(fjz0,tz);
 750
 751             fjptrA             = f+j_coord_offsetA;
 752             fjptrB             = f+j_coord_offsetB;
 753             fjptrC             = f+j_coord_offsetC;
 754             fjptrD             = f+j_coord_offsetD;
 755
 756             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 757
 758             /* Inner loop uses 81 flops */
 759         }
 760
 761         if(jidx<j_index_end)
 762         {
 763
 764             /* Get j neighbor index, and coordinate index */
 765             jnrlistA         = jjnr[jidx];
 766             jnrlistB         = jjnr[jidx+1];
 767             jnrlistC         = jjnr[jidx+2];
 768             jnrlistD         = jjnr[jidx+3];
 769             /* Sign of each element will be negative for non-real atoms.
 770              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 771              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 772              */
 773             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 774             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 775             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 776             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 777             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 778             j_coord_offsetA  = DIM*jnrA;
 779             j_coord_offsetB  = DIM*jnrB;
 780             j_coord_offsetC  = DIM*jnrC;
 781             j_coord_offsetD  = DIM*jnrD;
 782
 783             /* load j atom coordinates */
 784             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 785                                               x+j_coord_offsetC,x+j_coord_offsetD,
 786                                               &jx0,&jy0,&jz0);
 787
 788             /* Calculate displacement vector */
 789             dx00             = _mm_sub_ps(ix0,jx0);
 790             dy00             = _mm_sub_ps(iy0,jy0);
 791             dz00             = _mm_sub_ps(iz0,jz0);
 792             dx10             = _mm_sub_ps(ix1,jx0);
 793             dy10             = _mm_sub_ps(iy1,jy0);
 794             dz10             = _mm_sub_ps(iz1,jz0);
 795             dx20             = _mm_sub_ps(ix2,jx0);
 796             dy20             = _mm_sub_ps(iy2,jy0);
 797             dz20             = _mm_sub_ps(iz2,jz0);
 798
 799             /* Calculate squared distance and things based on it */
 800             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 801             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 802             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 803
 804             rinv00           = sse2_invsqrt_f(rsq00);
 805             rinv10           = sse2_invsqrt_f(rsq10);
 806             rinv20           = sse2_invsqrt_f(rsq20);
 807
 808             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 809             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 810             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 811
 812             /* Load parameters for j particles */
 813             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 814                                                               charge+jnrC+0,charge+jnrD+0);
 815
 816             fjx0             = _mm_setzero_ps();
 817             fjy0             = _mm_setzero_ps();
 818             fjz0             = _mm_setzero_ps();
 819
 820             /**************************
 821              * CALCULATE INTERACTIONS *
 822              **************************/
 823
 824             /* Compute parameters for interactions between i and j atoms */
 825             qq00             = _mm_mul_ps(iq0,jq0);
 826
 827             /* COULOMB ELECTROSTATICS */
 828             velec            = _mm_mul_ps(qq00,rinv00);
 829             felec            = _mm_mul_ps(velec,rinvsq00);
 830
 831             fscal            = felec;
 832
 833             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 834
 835             /* Calculate temporary vectorial force */
 836             tx               = _mm_mul_ps(fscal,dx00);
 837             ty               = _mm_mul_ps(fscal,dy00);
 838             tz               = _mm_mul_ps(fscal,dz00);
 839
 840             /* Update vectorial force */
 841             fix0             = _mm_add_ps(fix0,tx);
 842             fiy0             = _mm_add_ps(fiy0,ty);
 843             fiz0             = _mm_add_ps(fiz0,tz);
 844
 845             fjx0             = _mm_add_ps(fjx0,tx);
 846             fjy0             = _mm_add_ps(fjy0,ty);
 847             fjz0             = _mm_add_ps(fjz0,tz);
 848
 849             /**************************
 850              * CALCULATE INTERACTIONS *
 851              **************************/
 852
 853             /* Compute parameters for interactions between i and j atoms */
 854             qq10             = _mm_mul_ps(iq1,jq0);
 855
 856             /* COULOMB ELECTROSTATICS */
 857             velec            = _mm_mul_ps(qq10,rinv10);
 858             felec            = _mm_mul_ps(velec,rinvsq10);
 859
 860             fscal            = felec;
 861
 862             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 863
 864             /* Calculate temporary vectorial force */
 865             tx               = _mm_mul_ps(fscal,dx10);
 866             ty               = _mm_mul_ps(fscal,dy10);
 867             tz               = _mm_mul_ps(fscal,dz10);
 868
 869             /* Update vectorial force */
 870             fix1             = _mm_add_ps(fix1,tx);
 871             fiy1             = _mm_add_ps(fiy1,ty);
 872             fiz1             = _mm_add_ps(fiz1,tz);
 873
 874             fjx0             = _mm_add_ps(fjx0,tx);
 875             fjy0             = _mm_add_ps(fjy0,ty);
 876             fjz0             = _mm_add_ps(fjz0,tz);
 877
 878             /**************************
 879              * CALCULATE INTERACTIONS *
 880              **************************/
 881
 882             /* Compute parameters for interactions between i and j atoms */
 883             qq20             = _mm_mul_ps(iq2,jq0);
 884
 885             /* COULOMB ELECTROSTATICS */
 886             velec            = _mm_mul_ps(qq20,rinv20);
 887             felec            = _mm_mul_ps(velec,rinvsq20);
 888
 889             fscal            = felec;
 890
 891             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 892
 893             /* Calculate temporary vectorial force */
 894             tx               = _mm_mul_ps(fscal,dx20);
 895             ty               = _mm_mul_ps(fscal,dy20);
 896             tz               = _mm_mul_ps(fscal,dz20);
 897
 898             /* Update vectorial force */
 899             fix2             = _mm_add_ps(fix2,tx);
 900             fiy2             = _mm_add_ps(fiy2,ty);
 901             fiz2             = _mm_add_ps(fiz2,tz);
 902
 903             fjx0             = _mm_add_ps(fjx0,tx);
 904             fjy0             = _mm_add_ps(fjy0,ty);
 905             fjz0             = _mm_add_ps(fjz0,tz);
 906
 907             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 908             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 909             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 910             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 911
 912             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 913
 914             /* Inner loop uses 81 flops */
 915         }
 916
 917         /* End of innermost loop */
 918
 919         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 920                                               f+i_coord_offset,fshift+i_shift_offset);
 921
 922         /* Increment number of inner iterations */
 923         inneriter                  += j_index_end - j_index_start;
 924
 925         /* Outer loop uses 18 flops */
 926     }
 927
 928     /* Increment number of outer iterations */
 929     outeriter        += nri;
 930
 931     /* Update outer/inner flops */
 932
 933     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*81);
 934 }