src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecCoul_VdwLJ_GeomW3P1_sse4_1_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, 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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_double.h"
  48 #include "kernelutil_x86_sse4_1_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW3P1_VF_sse4_1_double
  52  * Electrostatics interaction: Coulomb
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_VF_sse4_1_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwjidx0A,vdwjidx0B;
  87     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  89     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  90     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  91     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     int              nvdwtype;
  94     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  95     int              *vdwtype;
  96     real             *vdwparam;
  97     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  98     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  99     __m128d          dummy_mask,cutoff_mask;
 100     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 101     __m128d          one     = _mm_set1_pd(1.0);
 102     __m128d          two     = _mm_set1_pd(2.0);
 103     x                = xx[0];
 104     f                = ff[0];
 105
 106     nri              = nlist->nri;
 107     iinr             = nlist->iinr;
 108     jindex           = nlist->jindex;
 109     jjnr             = nlist->jjnr;
 110     shiftidx         = nlist->shift;
 111     gid              = nlist->gid;
 112     shiftvec         = fr->shift_vec[0];
 113     fshift           = fr->fshift[0];
 114     facel            = _mm_set1_pd(fr->epsfac);
 115     charge           = mdatoms->chargeA;
 116     nvdwtype         = fr->ntype;
 117     vdwparam         = fr->nbfp;
 118     vdwtype          = mdatoms->typeA;
 119
 120     /* Setup water-specific parameters */
 121     inr              = nlist->iinr[0];
 122     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 123     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 124     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 125     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 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         vvdwsum          = _mm_setzero_pd();
 166
 167         /* Start inner kernel loop */
 168         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 169         {
 170
 171             /* Get j neighbor index, and coordinate index */
 172             jnrA             = jjnr[jidx];
 173             jnrB             = jjnr[jidx+1];
 174             j_coord_offsetA  = DIM*jnrA;
 175             j_coord_offsetB  = DIM*jnrB;
 176
 177             /* load j atom coordinates */
 178             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 179                                               &jx0,&jy0,&jz0);
 180
 181             /* Calculate displacement vector */
 182             dx00             = _mm_sub_pd(ix0,jx0);
 183             dy00             = _mm_sub_pd(iy0,jy0);
 184             dz00             = _mm_sub_pd(iz0,jz0);
 185             dx10             = _mm_sub_pd(ix1,jx0);
 186             dy10             = _mm_sub_pd(iy1,jy0);
 187             dz10             = _mm_sub_pd(iz1,jz0);
 188             dx20             = _mm_sub_pd(ix2,jx0);
 189             dy20             = _mm_sub_pd(iy2,jy0);
 190             dz20             = _mm_sub_pd(iz2,jz0);
 191
 192             /* Calculate squared distance and things based on it */
 193             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 194             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 195             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 196
 197             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 198             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 199             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 200
 201             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 202             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 203             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 204
 205             /* Load parameters for j particles */
 206             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 207             vdwjidx0A        = 2*vdwtype[jnrA+0];
 208             vdwjidx0B        = 2*vdwtype[jnrB+0];
 209
 210             fjx0             = _mm_setzero_pd();
 211             fjy0             = _mm_setzero_pd();
 212             fjz0             = _mm_setzero_pd();
 213
 214             /**************************
 215              * CALCULATE INTERACTIONS *
 216              **************************/
 217
 218             /* Compute parameters for interactions between i and j atoms */
 219             qq00             = _mm_mul_pd(iq0,jq0);
 220             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 221                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 222
 223             /* COULOMB ELECTROSTATICS */
 224             velec            = _mm_mul_pd(qq00,rinv00);
 225             felec            = _mm_mul_pd(velec,rinvsq00);
 226
 227             /* LENNARD-JONES DISPERSION/REPULSION */
 228
 229             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 230             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 231             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 232             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 233             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 234
 235             /* Update potential sum for this i atom from the interaction with this j atom. */
 236             velecsum         = _mm_add_pd(velecsum,velec);
 237             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 238
 239             fscal            = _mm_add_pd(felec,fvdw);
 240
 241             /* Calculate temporary vectorial force */
 242             tx               = _mm_mul_pd(fscal,dx00);
 243             ty               = _mm_mul_pd(fscal,dy00);
 244             tz               = _mm_mul_pd(fscal,dz00);
 245
 246             /* Update vectorial force */
 247             fix0             = _mm_add_pd(fix0,tx);
 248             fiy0             = _mm_add_pd(fiy0,ty);
 249             fiz0             = _mm_add_pd(fiz0,tz);
 250
 251             fjx0             = _mm_add_pd(fjx0,tx);
 252             fjy0             = _mm_add_pd(fjy0,ty);
 253             fjz0             = _mm_add_pd(fjz0,tz);
 254
 255             /**************************
 256              * CALCULATE INTERACTIONS *
 257              **************************/
 258
 259             /* Compute parameters for interactions between i and j atoms */
 260             qq10             = _mm_mul_pd(iq1,jq0);
 261
 262             /* COULOMB ELECTROSTATICS */
 263             velec            = _mm_mul_pd(qq10,rinv10);
 264             felec            = _mm_mul_pd(velec,rinvsq10);
 265
 266             /* Update potential sum for this i atom from the interaction with this j atom. */
 267             velecsum         = _mm_add_pd(velecsum,velec);
 268
 269             fscal            = felec;
 270
 271             /* Calculate temporary vectorial force */
 272             tx               = _mm_mul_pd(fscal,dx10);
 273             ty               = _mm_mul_pd(fscal,dy10);
 274             tz               = _mm_mul_pd(fscal,dz10);
 275
 276             /* Update vectorial force */
 277             fix1             = _mm_add_pd(fix1,tx);
 278             fiy1             = _mm_add_pd(fiy1,ty);
 279             fiz1             = _mm_add_pd(fiz1,tz);
 280
 281             fjx0             = _mm_add_pd(fjx0,tx);
 282             fjy0             = _mm_add_pd(fjy0,ty);
 283             fjz0             = _mm_add_pd(fjz0,tz);
 284
 285             /**************************
 286              * CALCULATE INTERACTIONS *
 287              **************************/
 288
 289             /* Compute parameters for interactions between i and j atoms */
 290             qq20             = _mm_mul_pd(iq2,jq0);
 291
 292             /* COULOMB ELECTROSTATICS */
 293             velec            = _mm_mul_pd(qq20,rinv20);
 294             felec            = _mm_mul_pd(velec,rinvsq20);
 295
 296             /* Update potential sum for this i atom from the interaction with this j atom. */
 297             velecsum         = _mm_add_pd(velecsum,velec);
 298
 299             fscal            = felec;
 300
 301             /* Calculate temporary vectorial force */
 302             tx               = _mm_mul_pd(fscal,dx20);
 303             ty               = _mm_mul_pd(fscal,dy20);
 304             tz               = _mm_mul_pd(fscal,dz20);
 305
 306             /* Update vectorial force */
 307             fix2             = _mm_add_pd(fix2,tx);
 308             fiy2             = _mm_add_pd(fiy2,ty);
 309             fiz2             = _mm_add_pd(fiz2,tz);
 310
 311             fjx0             = _mm_add_pd(fjx0,tx);
 312             fjy0             = _mm_add_pd(fjy0,ty);
 313             fjz0             = _mm_add_pd(fjz0,tz);
 314
 315             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 316
 317             /* Inner loop uses 99 flops */
 318         }
 319
 320         if(jidx<j_index_end)
 321         {
 322
 323             jnrA             = jjnr[jidx];
 324             j_coord_offsetA  = DIM*jnrA;
 325
 326             /* load j atom coordinates */
 327             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 328                                               &jx0,&jy0,&jz0);
 329
 330             /* Calculate displacement vector */
 331             dx00             = _mm_sub_pd(ix0,jx0);
 332             dy00             = _mm_sub_pd(iy0,jy0);
 333             dz00             = _mm_sub_pd(iz0,jz0);
 334             dx10             = _mm_sub_pd(ix1,jx0);
 335             dy10             = _mm_sub_pd(iy1,jy0);
 336             dz10             = _mm_sub_pd(iz1,jz0);
 337             dx20             = _mm_sub_pd(ix2,jx0);
 338             dy20             = _mm_sub_pd(iy2,jy0);
 339             dz20             = _mm_sub_pd(iz2,jz0);
 340
 341             /* Calculate squared distance and things based on it */
 342             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 343             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 344             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 345
 346             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 347             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 348             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 349
 350             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 351             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 352             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 353
 354             /* Load parameters for j particles */
 355             jq0              = _mm_load_sd(charge+jnrA+0);
 356             vdwjidx0A        = 2*vdwtype[jnrA+0];
 357
 358             fjx0             = _mm_setzero_pd();
 359             fjy0             = _mm_setzero_pd();
 360             fjz0             = _mm_setzero_pd();
 361
 362             /**************************
 363              * CALCULATE INTERACTIONS *
 364              **************************/
 365
 366             /* Compute parameters for interactions between i and j atoms */
 367             qq00             = _mm_mul_pd(iq0,jq0);
 368             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 369
 370             /* COULOMB ELECTROSTATICS */
 371             velec            = _mm_mul_pd(qq00,rinv00);
 372             felec            = _mm_mul_pd(velec,rinvsq00);
 373
 374             /* LENNARD-JONES DISPERSION/REPULSION */
 375
 376             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 377             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 378             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 379             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 380             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 381
 382             /* Update potential sum for this i atom from the interaction with this j atom. */
 383             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 384             velecsum         = _mm_add_pd(velecsum,velec);
 385             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 386             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 387
 388             fscal            = _mm_add_pd(felec,fvdw);
 389
 390             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 391
 392             /* Calculate temporary vectorial force */
 393             tx               = _mm_mul_pd(fscal,dx00);
 394             ty               = _mm_mul_pd(fscal,dy00);
 395             tz               = _mm_mul_pd(fscal,dz00);
 396
 397             /* Update vectorial force */
 398             fix0             = _mm_add_pd(fix0,tx);
 399             fiy0             = _mm_add_pd(fiy0,ty);
 400             fiz0             = _mm_add_pd(fiz0,tz);
 401
 402             fjx0             = _mm_add_pd(fjx0,tx);
 403             fjy0             = _mm_add_pd(fjy0,ty);
 404             fjz0             = _mm_add_pd(fjz0,tz);
 405
 406             /**************************
 407              * CALCULATE INTERACTIONS *
 408              **************************/
 409
 410             /* Compute parameters for interactions between i and j atoms */
 411             qq10             = _mm_mul_pd(iq1,jq0);
 412
 413             /* COULOMB ELECTROSTATICS */
 414             velec            = _mm_mul_pd(qq10,rinv10);
 415             felec            = _mm_mul_pd(velec,rinvsq10);
 416
 417             /* Update potential sum for this i atom from the interaction with this j atom. */
 418             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 419             velecsum         = _mm_add_pd(velecsum,velec);
 420
 421             fscal            = felec;
 422
 423             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 424
 425             /* Calculate temporary vectorial force */
 426             tx               = _mm_mul_pd(fscal,dx10);
 427             ty               = _mm_mul_pd(fscal,dy10);
 428             tz               = _mm_mul_pd(fscal,dz10);
 429
 430             /* Update vectorial force */
 431             fix1             = _mm_add_pd(fix1,tx);
 432             fiy1             = _mm_add_pd(fiy1,ty);
 433             fiz1             = _mm_add_pd(fiz1,tz);
 434
 435             fjx0             = _mm_add_pd(fjx0,tx);
 436             fjy0             = _mm_add_pd(fjy0,ty);
 437             fjz0             = _mm_add_pd(fjz0,tz);
 438
 439             /**************************
 440              * CALCULATE INTERACTIONS *
 441              **************************/
 442
 443             /* Compute parameters for interactions between i and j atoms */
 444             qq20             = _mm_mul_pd(iq2,jq0);
 445
 446             /* COULOMB ELECTROSTATICS */
 447             velec            = _mm_mul_pd(qq20,rinv20);
 448             felec            = _mm_mul_pd(velec,rinvsq20);
 449
 450             /* Update potential sum for this i atom from the interaction with this j atom. */
 451             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 452             velecsum         = _mm_add_pd(velecsum,velec);
 453
 454             fscal            = felec;
 455
 456             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 457
 458             /* Calculate temporary vectorial force */
 459             tx               = _mm_mul_pd(fscal,dx20);
 460             ty               = _mm_mul_pd(fscal,dy20);
 461             tz               = _mm_mul_pd(fscal,dz20);
 462
 463             /* Update vectorial force */
 464             fix2             = _mm_add_pd(fix2,tx);
 465             fiy2             = _mm_add_pd(fiy2,ty);
 466             fiz2             = _mm_add_pd(fiz2,tz);
 467
 468             fjx0             = _mm_add_pd(fjx0,tx);
 469             fjy0             = _mm_add_pd(fjy0,ty);
 470             fjz0             = _mm_add_pd(fjz0,tz);
 471
 472             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 473
 474             /* Inner loop uses 99 flops */
 475         }
 476
 477         /* End of innermost loop */
 478
 479         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 480                                               f+i_coord_offset,fshift+i_shift_offset);
 481
 482         ggid                        = gid[iidx];
 483         /* Update potential energies */
 484         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 485         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 486
 487         /* Increment number of inner iterations */
 488         inneriter                  += j_index_end - j_index_start;
 489
 490         /* Outer loop uses 20 flops */
 491     }
 492
 493     /* Increment number of outer iterations */
 494     outeriter        += nri;
 495
 496     /* Update outer/inner flops */
 497
 498     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*99);
 499 }
 500 /*
 501  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_sse4_1_double
 502  * Electrostatics interaction: Coulomb
 503  * VdW interaction:            LennardJones
 504  * Geometry:                   Water3-Particle
 505  * Calculate force/pot:        Force
 506  */
 507 void
 508 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_sse4_1_double
 509                     (t_nblist                    * gmx_restrict       nlist,
 510                      rvec                        * gmx_restrict          xx,
 511                      rvec                        * gmx_restrict          ff,
 512                      t_forcerec                  * gmx_restrict          fr,
 513                      t_mdatoms                   * gmx_restrict     mdatoms,
 514                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 515                      t_nrnb                      * gmx_restrict        nrnb)
 516 {
 517     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 518      * just 0 for non-waters.
 519      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 520      * jnr indices corresponding to data put in the four positions in the SIMD register.
 521      */
 522     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 523     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 524     int              jnrA,jnrB;
 525     int              j_coord_offsetA,j_coord_offsetB;
 526     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 527     real             rcutoff_scalar;
 528     real             *shiftvec,*fshift,*x,*f;
 529     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 530     int              vdwioffset0;
 531     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 532     int              vdwioffset1;
 533     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 534     int              vdwioffset2;
 535     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 536     int              vdwjidx0A,vdwjidx0B;
 537     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 538     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 539     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 540     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 541     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 542     real             *charge;
 543     int              nvdwtype;
 544     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 545     int              *vdwtype;
 546     real             *vdwparam;
 547     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 548     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 549     __m128d          dummy_mask,cutoff_mask;
 550     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 551     __m128d          one     = _mm_set1_pd(1.0);
 552     __m128d          two     = _mm_set1_pd(2.0);
 553     x                = xx[0];
 554     f                = ff[0];
 555
 556     nri              = nlist->nri;
 557     iinr             = nlist->iinr;
 558     jindex           = nlist->jindex;
 559     jjnr             = nlist->jjnr;
 560     shiftidx         = nlist->shift;
 561     gid              = nlist->gid;
 562     shiftvec         = fr->shift_vec[0];
 563     fshift           = fr->fshift[0];
 564     facel            = _mm_set1_pd(fr->epsfac);
 565     charge           = mdatoms->chargeA;
 566     nvdwtype         = fr->ntype;
 567     vdwparam         = fr->nbfp;
 568     vdwtype          = mdatoms->typeA;
 569
 570     /* Setup water-specific parameters */
 571     inr              = nlist->iinr[0];
 572     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 573     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 574     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 575     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 576
 577     /* Avoid stupid compiler warnings */
 578     jnrA = jnrB = 0;
 579     j_coord_offsetA = 0;
 580     j_coord_offsetB = 0;
 581
 582     outeriter        = 0;
 583     inneriter        = 0;
 584
 585     /* Start outer loop over neighborlists */
 586     for(iidx=0; iidx<nri; iidx++)
 587     {
 588         /* Load shift vector for this list */
 589         i_shift_offset   = DIM*shiftidx[iidx];
 590
 591         /* Load limits for loop over neighbors */
 592         j_index_start    = jindex[iidx];
 593         j_index_end      = jindex[iidx+1];
 594
 595         /* Get outer coordinate index */
 596         inr              = iinr[iidx];
 597         i_coord_offset   = DIM*inr;
 598
 599         /* Load i particle coords and add shift vector */
 600         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 601                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 602
 603         fix0             = _mm_setzero_pd();
 604         fiy0             = _mm_setzero_pd();
 605         fiz0             = _mm_setzero_pd();
 606         fix1             = _mm_setzero_pd();
 607         fiy1             = _mm_setzero_pd();
 608         fiz1             = _mm_setzero_pd();
 609         fix2             = _mm_setzero_pd();
 610         fiy2             = _mm_setzero_pd();
 611         fiz2             = _mm_setzero_pd();
 612
 613         /* Start inner kernel loop */
 614         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 615         {
 616
 617             /* Get j neighbor index, and coordinate index */
 618             jnrA             = jjnr[jidx];
 619             jnrB             = jjnr[jidx+1];
 620             j_coord_offsetA  = DIM*jnrA;
 621             j_coord_offsetB  = DIM*jnrB;
 622
 623             /* load j atom coordinates */
 624             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 625                                               &jx0,&jy0,&jz0);
 626
 627             /* Calculate displacement vector */
 628             dx00             = _mm_sub_pd(ix0,jx0);
 629             dy00             = _mm_sub_pd(iy0,jy0);
 630             dz00             = _mm_sub_pd(iz0,jz0);
 631             dx10             = _mm_sub_pd(ix1,jx0);
 632             dy10             = _mm_sub_pd(iy1,jy0);
 633             dz10             = _mm_sub_pd(iz1,jz0);
 634             dx20             = _mm_sub_pd(ix2,jx0);
 635             dy20             = _mm_sub_pd(iy2,jy0);
 636             dz20             = _mm_sub_pd(iz2,jz0);
 637
 638             /* Calculate squared distance and things based on it */
 639             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 640             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 641             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 642
 643             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 644             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 645             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 646
 647             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 648             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 649             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 650
 651             /* Load parameters for j particles */
 652             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 653             vdwjidx0A        = 2*vdwtype[jnrA+0];
 654             vdwjidx0B        = 2*vdwtype[jnrB+0];
 655
 656             fjx0             = _mm_setzero_pd();
 657             fjy0             = _mm_setzero_pd();
 658             fjz0             = _mm_setzero_pd();
 659
 660             /**************************
 661              * CALCULATE INTERACTIONS *
 662              **************************/
 663
 664             /* Compute parameters for interactions between i and j atoms */
 665             qq00             = _mm_mul_pd(iq0,jq0);
 666             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 667                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 668
 669             /* COULOMB ELECTROSTATICS */
 670             velec            = _mm_mul_pd(qq00,rinv00);
 671             felec            = _mm_mul_pd(velec,rinvsq00);
 672
 673             /* LENNARD-JONES DISPERSION/REPULSION */
 674
 675             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 676             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 677
 678             fscal            = _mm_add_pd(felec,fvdw);
 679
 680             /* Calculate temporary vectorial force */
 681             tx               = _mm_mul_pd(fscal,dx00);
 682             ty               = _mm_mul_pd(fscal,dy00);
 683             tz               = _mm_mul_pd(fscal,dz00);
 684
 685             /* Update vectorial force */
 686             fix0             = _mm_add_pd(fix0,tx);
 687             fiy0             = _mm_add_pd(fiy0,ty);
 688             fiz0             = _mm_add_pd(fiz0,tz);
 689
 690             fjx0             = _mm_add_pd(fjx0,tx);
 691             fjy0             = _mm_add_pd(fjy0,ty);
 692             fjz0             = _mm_add_pd(fjz0,tz);
 693
 694             /**************************
 695              * CALCULATE INTERACTIONS *
 696              **************************/
 697
 698             /* Compute parameters for interactions between i and j atoms */
 699             qq10             = _mm_mul_pd(iq1,jq0);
 700
 701             /* COULOMB ELECTROSTATICS */
 702             velec            = _mm_mul_pd(qq10,rinv10);
 703             felec            = _mm_mul_pd(velec,rinvsq10);
 704
 705             fscal            = felec;
 706
 707             /* Calculate temporary vectorial force */
 708             tx               = _mm_mul_pd(fscal,dx10);
 709             ty               = _mm_mul_pd(fscal,dy10);
 710             tz               = _mm_mul_pd(fscal,dz10);
 711
 712             /* Update vectorial force */
 713             fix1             = _mm_add_pd(fix1,tx);
 714             fiy1             = _mm_add_pd(fiy1,ty);
 715             fiz1             = _mm_add_pd(fiz1,tz);
 716
 717             fjx0             = _mm_add_pd(fjx0,tx);
 718             fjy0             = _mm_add_pd(fjy0,ty);
 719             fjz0             = _mm_add_pd(fjz0,tz);
 720
 721             /**************************
 722              * CALCULATE INTERACTIONS *
 723              **************************/
 724
 725             /* Compute parameters for interactions between i and j atoms */
 726             qq20             = _mm_mul_pd(iq2,jq0);
 727
 728             /* COULOMB ELECTROSTATICS */
 729             velec            = _mm_mul_pd(qq20,rinv20);
 730             felec            = _mm_mul_pd(velec,rinvsq20);
 731
 732             fscal            = felec;
 733
 734             /* Calculate temporary vectorial force */
 735             tx               = _mm_mul_pd(fscal,dx20);
 736             ty               = _mm_mul_pd(fscal,dy20);
 737             tz               = _mm_mul_pd(fscal,dz20);
 738
 739             /* Update vectorial force */
 740             fix2             = _mm_add_pd(fix2,tx);
 741             fiy2             = _mm_add_pd(fiy2,ty);
 742             fiz2             = _mm_add_pd(fiz2,tz);
 743
 744             fjx0             = _mm_add_pd(fjx0,tx);
 745             fjy0             = _mm_add_pd(fjy0,ty);
 746             fjz0             = _mm_add_pd(fjz0,tz);
 747
 748             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 749
 750             /* Inner loop uses 91 flops */
 751         }
 752
 753         if(jidx<j_index_end)
 754         {
 755
 756             jnrA             = jjnr[jidx];
 757             j_coord_offsetA  = DIM*jnrA;
 758
 759             /* load j atom coordinates */
 760             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 761                                               &jx0,&jy0,&jz0);
 762
 763             /* Calculate displacement vector */
 764             dx00             = _mm_sub_pd(ix0,jx0);
 765             dy00             = _mm_sub_pd(iy0,jy0);
 766             dz00             = _mm_sub_pd(iz0,jz0);
 767             dx10             = _mm_sub_pd(ix1,jx0);
 768             dy10             = _mm_sub_pd(iy1,jy0);
 769             dz10             = _mm_sub_pd(iz1,jz0);
 770             dx20             = _mm_sub_pd(ix2,jx0);
 771             dy20             = _mm_sub_pd(iy2,jy0);
 772             dz20             = _mm_sub_pd(iz2,jz0);
 773
 774             /* Calculate squared distance and things based on it */
 775             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 776             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 777             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 778
 779             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 780             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 781             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 782
 783             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 784             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 785             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 786
 787             /* Load parameters for j particles */
 788             jq0              = _mm_load_sd(charge+jnrA+0);
 789             vdwjidx0A        = 2*vdwtype[jnrA+0];
 790
 791             fjx0             = _mm_setzero_pd();
 792             fjy0             = _mm_setzero_pd();
 793             fjz0             = _mm_setzero_pd();
 794
 795             /**************************
 796              * CALCULATE INTERACTIONS *
 797              **************************/
 798
 799             /* Compute parameters for interactions between i and j atoms */
 800             qq00             = _mm_mul_pd(iq0,jq0);
 801             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 802
 803             /* COULOMB ELECTROSTATICS */
 804             velec            = _mm_mul_pd(qq00,rinv00);
 805             felec            = _mm_mul_pd(velec,rinvsq00);
 806
 807             /* LENNARD-JONES DISPERSION/REPULSION */
 808
 809             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 810             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 811
 812             fscal            = _mm_add_pd(felec,fvdw);
 813
 814             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 815
 816             /* Calculate temporary vectorial force */
 817             tx               = _mm_mul_pd(fscal,dx00);
 818             ty               = _mm_mul_pd(fscal,dy00);
 819             tz               = _mm_mul_pd(fscal,dz00);
 820
 821             /* Update vectorial force */
 822             fix0             = _mm_add_pd(fix0,tx);
 823             fiy0             = _mm_add_pd(fiy0,ty);
 824             fiz0             = _mm_add_pd(fiz0,tz);
 825
 826             fjx0             = _mm_add_pd(fjx0,tx);
 827             fjy0             = _mm_add_pd(fjy0,ty);
 828             fjz0             = _mm_add_pd(fjz0,tz);
 829
 830             /**************************
 831              * CALCULATE INTERACTIONS *
 832              **************************/
 833
 834             /* Compute parameters for interactions between i and j atoms */
 835             qq10             = _mm_mul_pd(iq1,jq0);
 836
 837             /* COULOMB ELECTROSTATICS */
 838             velec            = _mm_mul_pd(qq10,rinv10);
 839             felec            = _mm_mul_pd(velec,rinvsq10);
 840
 841             fscal            = felec;
 842
 843             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 844
 845             /* Calculate temporary vectorial force */
 846             tx               = _mm_mul_pd(fscal,dx10);
 847             ty               = _mm_mul_pd(fscal,dy10);
 848             tz               = _mm_mul_pd(fscal,dz10);
 849
 850             /* Update vectorial force */
 851             fix1             = _mm_add_pd(fix1,tx);
 852             fiy1             = _mm_add_pd(fiy1,ty);
 853             fiz1             = _mm_add_pd(fiz1,tz);
 854
 855             fjx0             = _mm_add_pd(fjx0,tx);
 856             fjy0             = _mm_add_pd(fjy0,ty);
 857             fjz0             = _mm_add_pd(fjz0,tz);
 858
 859             /**************************
 860              * CALCULATE INTERACTIONS *
 861              **************************/
 862
 863             /* Compute parameters for interactions between i and j atoms */
 864             qq20             = _mm_mul_pd(iq2,jq0);
 865
 866             /* COULOMB ELECTROSTATICS */
 867             velec            = _mm_mul_pd(qq20,rinv20);
 868             felec            = _mm_mul_pd(velec,rinvsq20);
 869
 870             fscal            = felec;
 871
 872             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 873
 874             /* Calculate temporary vectorial force */
 875             tx               = _mm_mul_pd(fscal,dx20);
 876             ty               = _mm_mul_pd(fscal,dy20);
 877             tz               = _mm_mul_pd(fscal,dz20);
 878
 879             /* Update vectorial force */
 880             fix2             = _mm_add_pd(fix2,tx);
 881             fiy2             = _mm_add_pd(fiy2,ty);
 882             fiz2             = _mm_add_pd(fiz2,tz);
 883
 884             fjx0             = _mm_add_pd(fjx0,tx);
 885             fjy0             = _mm_add_pd(fjy0,ty);
 886             fjz0             = _mm_add_pd(fjz0,tz);
 887
 888             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 889
 890             /* Inner loop uses 91 flops */
 891         }
 892
 893         /* End of innermost loop */
 894
 895         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 896                                               f+i_coord_offset,fshift+i_shift_offset);
 897
 898         /* Increment number of inner iterations */
 899         inneriter                  += j_index_end - j_index_start;
 900
 901         /* Outer loop uses 18 flops */
 902     }
 903
 904     /* Increment number of outer iterations */
 905     outeriter        += nri;
 906
 907     /* Update outer/inner flops */
 908
 909     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*91);
 910 }